It keeps the design simpler to understand, and is probably more efficient

Just need to write the SQL queries now

This halves the 'time_elapsed_opening' metric, and gives a ~30% perf
improvement.

This was a waste of time, because Datafusion only uses FileFormatFactory when opening
a single file

it appears calling the backend provenance's `whereare` without any
kind of heritage or so does not help much...

Make it possible to enable or disable filtering license files via a
`filter_licenses` config option.

Also add a few more license files in the list of know license swhids.

This filters (return None) provenance results for a bunch of well-known
content and directory objects (empty file etc.) so that provenance
queries on these objects do not hammer swh-graph with these known
objects.

This is far from a definitive solution, but should help for now.

5011 is already in use.

It enables to benefit from provenance type checking using mypy
in other swh components.

It enables to use the swh_provenance fixture in other swh components.

Also:
- add a few pre-commit configs
- normalize the tox file
- fix the docs/Makefile

There is only rust left in swh-graph now.