They are not of interest anymore as swh-web is the only client and now
directly uses functions defined in swh.storage.algos.diff.

They got copied to test_postgresql.py instead of moved.

Also move the round_to_milliseconds function to the storage.utils module.

Fixes a regression in 038a219f,
as it made the converter expect a list.

When adding this column, we made it default to null instead of defaulting
to an empty array, so existing records were initialized will null.

This commit migrates these nulls to empty arrays, then adds a constraint
to enforce it in the future.

Fixes a regression in 038a219f,
as it made converters expect a boolean.

We stopped writing this kind of nulls since we started using model
objects for insertions, but didn't migrate existing data.

This commit migrates these nulls to false, then adds a constraint
to enforce it in the future.

This field wasn't backfilled; and it wasn't caught by tests, because
revisions in swh.journal.tests.journal_data.TEST_OBJECTS were had
an empty extra_header field.

Starting with swh-journal v0.4.3, one of the revisions in TEST_OBJECTS
has a non-empty extra_header field, so it will become a test failure
without this commit

As a temporary workaround to remain compatible with existing backfiller converters,
I made them convert back to dict before they are converted again to model objects.

This commit removes this workaround by making converters return model objects.

This is a change internal to the pg storage, that will be needed to make
revision_get, revision_log, and release_get return model objects.

in the two methods which need to switch between statements.

The method name building was done to shoehorn this statement switching
into the existing @_prepared_select_statement.

This introduces @_prepared_select_statements (plural), which does this
switching properly, using a dictionary.

It's not optional in the storage interface, so a None value can't be passed.

It's not used anywhere.

in_memory: Remove InMemoryStorage.skipped_content_* and implement InMemoryCqlRunner.skipped_content_*

For now this has no effect. However, in the near future, the CassandraStorage
will be in charge of some object types, so we need to merge objects
found in the CassandraStorage and those found directly in the InMemoryStorage.

For now it's only used for object counters; but future commits will
progressively move the in-memory's storage features to it.

This has no effect for now, other than deduplicating a method
and causing a name clash.

It will be used to implement the in-memory storage as a backend for the
cassandra storage.

They will be used by the in-mem implementation of CqlRunner.

cassandra: Make origin_visit_get_latest filter using any status of a visit, instead of just the last.

This fixes a mismatch in behavior with the pg and the in-mem storages

'SELECT *' should be avoided in prepared statements, see
https://docs.datastax.com/en/developer/java-driver/3.0/manual/statements/prepared/#avoid-preparing-select-queries

cassandra: add a TABLE class attribute to row classes, and use it to deduplicate prepared statements logic.

It will also be used in a future commit to generate 'select' prepared statements.

All methods of CqlRunner were decorated, which prevented mypy from doing
anything useful.

As I finally found a way to type the decorator (using
mypy_extensions.NamedArg), I can finally make mypy aware of the methods' types.

This commit (as well as all three of the last commits) also fixes issues found
by mypy thanks to this.