Previously the parse_external_definition function was returning a single
revision number regardless it was a revision specified with -rX or a peg
revision specified with @X.

However, the use and combination of these two parameters in the export
command from subversion can lead to different results
(see https://svnbook.red-bean.com/en/1.6/svn.advanced.pegrevs.html).

So ensure to extract both revision and peg revision in order to avoid
different behavior from the official subversion client when the loader
exports externals.

This fixes python 3.7 support due to poetry, a dependency of isort, that
removed support for that Python version in a recent release.

Subversion allow to specify the revision for an external definition as a
date instead of an integer identifier.

So when encountering such case, get the HEAD revision number for the
external at the specified date in order to export the correct version
of the files targeted by the external.

Related to #4727

It enables to get the HEAD revision number for a repository at a
specific date.

Related to #4727

It prevents "Remote access object already in use" errors.

Those are simply wrappers around functions from the converters module
are are nou used elsewhere.

Add some default parameters to SvnRepo class constructor in order to
simplify initialization of such object for standalone use.

Make origin_url parameter of info method optional.

Add some tests for the SvnRepo class.

This is a more meaninful name considering that module only contains
a single class named SvnRepo.

In order to remove warnings about /apidoc/*.rst files being included
multiple times in toc when building full swh documentation, prefer to
include module indices only when building standalone package documentation.

Also include them the proper sphinx way.

Related to T4496

Now that SvnRepo.propget supports URL as target, we can remove the
use of costly checkout operation and directly retrieve the whole
set of svn:externals properties.

This should greatly improve incremental loading of a big repository
in terms of performance.

subvertpy 0.11 has a buggy implementation of propget bindings when
target is an URL (https://github.com/jelmer/subvertpy/issues/35),
so as a workaround we implement propget for URL using non buggy
proplist bindings.

Retrying three times is enough as we use expontential backoff.

Previously the loader could be stuck more than twenty minutes in a row
when it encounters a dead external, now it would be a couple of minutes.

Copied directories might have externals so we also need to copy states
and update external paths in case externals list is later modified.

In order to detect all ascii characters that must be percent encoded
in svn URLs, add a brute force test and use urllib.parse.quote in
quote_svn_url function.

Such case can happen when an external definition is malformed.

Previously, the parsed malformed external was added to the directory state
with an empty external URL which could lead to unexpected side effects like
removing all previously exported valid externals.

Instead of maintaining file state based on svn properties across revisions
replay and trying to reconstruct the same file as with a svn export operation
after applying text deltas, prefer to simply export the file from the currently
processed revision when closing the associated file editor.

This greatly simplify the replay module implementation while approximatively
keeping the same performance as before.

Also add a test that would fail without these changes.

Related to T4673

When copying a directory from an ancestor revision, do not ignore
externals as properties are also copied by subversion so external
paths must also be exported.

In debug mode, when a hash tree computation divergence is detected after
replaying a revision, compute and display the diff between contents to
facilitate debugging of those type of issues.

Add more debug logs to the replay module to ease detection of issues.
Nevertheless, as those are quite verbose, only display them when setting
debug parameter of the loader to True.

When a tree computation divergence is detected after replaying a revision
add debug logs displaying the paths that differ or are missing between the
reconstructed repository filesystem and the exported one at that specific
revision.

It should help to gain some time when debugging such issues.

A subversion revision can contain new directories and files copied from
ancestor revisions but those were not perfectly handled in the commit
editor used to reconstruct the repository filesystem when replaying
revisions.

In particular previous implementation could not handle the case where a
path copied from an ancestor revision is replaced in a same commit (for
instance replacing a directory by a file with the same name).

These changes ensure that info about source path and source revision from
which a path is copied is passed to the commit editor methods as paramaters
in order to let them handle the copies but also that the replace operations
will be correctly replayed.

It also prevents OS error "Too many open files" when a really large files
tree is copied from an ancestor revision.

When dumping a subversion repository to file before loading it, compress
that file using gzip while producing it. It enables to save significant
disk space while dumping a large repository.

Also rework the way how truncated dump is handled now dump file is
compressed by providing the expected max revision number to be loaded
by svnadmin. If the number of loaded revisions matches, we can safely
continue the partial loading of the repository.

URLs provided as parameters to subvertpy.client.Client methods must
be quoted when it contains space characters or an assertion will be
raised by libsvn otherwise.

It exists some subtle cases in subversion repositories where external
paths defined on different directories can overlap so update replay
module implementation to handle those and avoid to erroneously remove
paths when replaying revisions.

When the "svnadmin load" command exits with error, report the svn
admin error in the ValueError exception raised by the function
init_svn_repo_from_dump.

This should help debugging those type of issues reported by sentry.

- pre-commit from 4.1.0 to 4.3.0,
- codespell from 2.2.1 to 2.2.2,
- black from 22.3.0 to 22.10.0 and
- flake8 from 4.0.1 to 5.0.4. Also freeze flake8 dependencies.

Also change flake8's repo config to github (the gitlab mirror
being outdated).

Use helper fixture loading_task_creation_for_listed_origin_test from
swh-loader-core and remove redundant tests.

Instead of maintaining a set of modified paths for each replayed revision,
use the swh.model.from_disk.Directory.collect method which performs the same
task by returning added or modified contents and directories since the last
collect operation.

Some subversion servers in the wild disabled anonymous access and require
to authenticate using a read only user account, typically with credentials
anonymous/anonymous.

So add support for basic authentication in the subversion loader, credentials
must be provided in the repository URL as with basic HTTP authentication.

An external definition can be of the following form (where XXX and YYY
are revision numbers):

	-r XXX <repo_url>@YYY

In that case, the official subversion client will export the revision
XXX of the external repository.

So ensure to have the same behavior when the subversion loader processes
a repository with such external defintion in it.

Temporary network failures can also happen when using subversion
remote access API so make single commit info retrieval operation
retryable.

This type of error comes from temporary network failure so the failed
svn operation can be retried.

The computed root directory name had a trailing slash which was making its
lookup in from_disk.Directory model fail and thus every archived revisions
were targetting the empty directory.

svnrdump does not handle repository URL redirection while svn client does.

So ensure to use redirected subversion origin URL to dump a repository.

Related to T3874

Using unnamed arguments is unsafe with celery tasks and parsing visit_date
is still required in case that parameter is used.

Recent changes in swh-scheduler add new parameters to the celery tasks
produced from swh.scheduler.model.ListedOrigin instances.

Those new parameters were not properly handled in the celery tasks
implementation for svn loading but also in some svn loader ones.

So ensure to handle any new parameters by not hardcoding the expected
ones in task signatures but also allowing any extra keyword parameters
in loader constructors.

Also remove explicit setting of visit_date if none has been provided
as task parameter as it is already handled in loader constructor.

Related to T4187