Add support for beancount v3

[LalitMaganti]

Switch to using beangulp which supports both v2 and v3.

Fixes: #253

[Nulifier]

This doesn't appear to work for the generic_importer_source.

I get the following error:

TypeError: expected str, bytes or os.PathLike object, not _FileMemo

This is related to the switch from _FileMemo to straight filepath strings. I managed to get past that by checking if the filepath has a name property and using that. When I got past that, I ran into the following issue:

TypeError: Importer.extract() got an unexpected keyword argument 'existing_entries' which is related to the change in call signature from the Importer.

[LalitMaganti]

Thanks for letting me know. 0c37757 should fix this.

[slashformotion]

LGTM

[Zburatorul]

@LalitMaganti thanks for kicking this off — I built on the same beangulp
approach and opened #263 as a more complete alternative. It takes the same beangulp approach as this PR (the generic_importer_source change is equivalent), but also carries what's needed for a complete v3 install:

• adds beangulp and beanquery to install_requires (this PR switches the
  imports but doesn't declare the new deps);
• migrates delete_transactions/rename_account off the removed
  beancount.query → beanquery;
• handles v3's __automatic__/__residual__ posting metadata in
  journal_editor;
• swaps pkg_resources → importlib.resources, fixes mypy under v3's
  py.typed, and updates CI to Python 3.9–3.12;
• keeps v2.3+ compatibility (beancount>=2.3.5, since beangulp/beanquery
  both support 2.3+).

CI is green across 3.9–3.12 on Linux/Windows/macOS and I validated it against a real ledger. Flagging here so we're not working the two in parallel — @jbms, happy to go with whichever you prefer.