fix(l2g): direct model path to hf repo when none and align training data filename

[ireneisdoomed]

✨ Context

This PR fixes 2 minor bugs that was causing the L2G prediction step to crash when model_path=None:

• L2GPrediction.from_credible_set defaults to the opentargets/locus_to_gene repo if model_path is none. Now, at the L2G step level this variable matches hf_repo_id as long as download_from_hub is set to true
• Small typo in the filename of the training set when exporting to HF

🛠 What does this PR implement

🙈 Missing

🚦 Before submitting

• Do these changes cover one single feature (one change at a time)?
• Did you read the contributor guideline?
• Did you make sure to update the documentation with your changes?
• Did you make sure there is no commented out code in this PR?
• Did you follow conventional commits standards in PR title and commit messages?
• Did you make sure the branch is up-to-date with the dev branch?
• Did you write any new necessary tests?
• Did you make sure the changes pass local tests (make test)?
• Did you make sure the changes pass pre-commit rules (e.g uv run pre-commit run --all-files)?

[project-defiant]

@ireneisdoomed do we have to make new release candidate to run the l2g steps ?

@javfg

[ireneisdoomed]

@ireneisdoomed do we have to make new release candidate to run the l2g steps ?

@javfg

Not necessarily but it would be the easiest.
The bug is not there if model_path is provided through the configuration (requires changes in orchestration)

[ireneisdoomed]

Train and predict train are now synced, the bug has been fixed.
In a following PR, I'll implement the training data sampling for a faster shapley calculation.