Refactor pyproject.toml and cleanup pyright restrictions

[mhusnain-tech]

This PR updates the pyproject.toml to align with recent changes in the repository. Main changes include:

• Cleaned up Pyright configuration by removing redundant 'false' flags for torch-related issues.
• Updated project metadata and classifiers for better consistency.
• Standardized dependency formatting.

This follows the work initiated in PR #1768 to handle Torch 2.5+ type-stub issues more cleanly.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (main@b546366). Learn more about missing BASE report.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1770   +/-   ##
=======================================
  Coverage        ?   86.75%           
=======================================
  Files           ?      141           
  Lines           ?    12317           
  Branches        ?        0           
=======================================
  Hits            ?    10685           
  Misses          ?     1632           
  Partials        ?        0           

Flag	Coverage Δ
fast	83.30% <ø> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

[manuelgloeckler]

Hey @mhusnain-tech ,

thanks for the suggestions. Yet, as visible in the CI. The current flags are necessary for newer torch versions.

This is due to errors like /home/runner/work/sbi/sbi/sbi/inference/trainers/npe/npe_c.py:289:32 - error: Operator "-" not supported for types "Tensor" and "Tensor | Module" appearing from mathematical operations with e.g. buffers. This is due to new torch stubs for type hints. Sure, some of these errors are indeed incomplete/imprecise typing on our end, but many are due to external imprecise typing which we cannot influence.

If you have expertise in pyright and know a better solution than the current one, open for any suggestions. But otherwise I would close this PR.

[manuelgloeckler]

Closing, as seemingly moved to #1773