Enable to check package documentation can be built without producing
sphinx warnings.

The sphinx environment is designed to be used in continuous integration
in order to prevent breaking documentation build when committing changes.

The sphinx-dev environment is designed to be used inside a full swh
development environment.

Related to T3258

This fixes the following warning:

<generated>:1: WARNING: Inline emphasis start-string without end-string.

Related to T2265

Right now we are generating two different binary mappings on disk for
the translation between SWHID <-> webgraph node ID:

1. The node2swh.bin reverse map, which contains a list of binary SWHID.
It allows O(1) access to the SWHID of a given node n by seek()ing to the
position (n * record size).

2. The swhid2node.bin map, which contains a list of <SWHID, node ID>
   ordered by SWHID. The node ID of a given SWHID can be found in
   O(log N) by doing a binary search on this list.

Because the swhid -> node binary search requires multiple seek() on the
disk, it is very slow and cannot be used for performance sensitive code.

The alternative route is to compute the node from the minimal perfect
hash function and then lookup the equivalent node in the permuted graph
using the .order file containing the permutation, which can be done in
O(1) and is extremely fast. However, this has two caveats:

- The .order file is ~150 GB, which would be too big to load in memory.

- MPH cannot check whether their input is valid. They could do so
  probabilistically if we signed them, but when replying to a query in
  the graph service, we want to be absolutely certain that a node is or
  is not present in the graph.

This code mitigates these problems in two ways. First, it memory-maps
the permutation file to avoid loading it in main memory. Second, it uses
a roundtrip check to detect invalid SWHIDs: we hash + permute the
SWHID, then use the reverse map to check that the obtained node ID is
associated to the original SWHID.

This is a big performance gain (basic benchmarks show a ~x3 speedup).
To go even faster, we offer a boolean option to skip the roundtrip
check, to use when we know that the input is always valid.

This will also allow us in the future to remove the swhid2node map
completely, however it is currently still in use by the Python frontend
to encode the SWHIDs. This will be done directly in the Java side in the
future.

tuning batch size by hand is no longer needed since 5a987aae

Rationale: we do not need to do so at all, because we are only interested in
commits that directly contain the content or dir (at some depth), nor in any of
their successors.

Note: this bug was not responsible for wrong answers in most cases (because
successors will tend to have higher timestamps), but incurred significant extra
time (4x in early benchmarks) due to exploring significant commit histories for
any file/dir appearing in large projects.

The SWHID class is deprecated, and no longer supports 'ori' objects.

Fix T2595

Since aiohttp 3.7, the internal handling of aiohttp.web.HTTPException has changed
and a aiohttp.web_response.Response is now constructed from the exception text
then returned.

aiohttp exception constructions in swh-graph were passing error message through
the body keyword parameter instead of the text one, leading to an error related
to unicode decoding as body is expected to be bytes.

Using the text keyword parameter when constructing aiohttp exceptions now ensures
that response body will be properly encoded.

Closes T2768

- Replace the old (and unused) .clang-format with Spotless
- Add a pre-commit hook to enforce style
- Fix style on the Java codebase