Conditionally clear MPS cache on OOM error

[tongchen126]

This commit will fix error on CUDA-only system.
Otherwise, the following error will occur:
"Cannot execute emptyCache() without MPS backend."

Issue

Please link the corresponding GitHub issue. If an issue does not already exist,
please open one to describe the bug or feature request before creating a pull request.

This allows us to discuss the proposal and helps avoid unnecessary work.

Motivation and Context

I encounterd error on my CUDA system. In torch, torch.backends.mps.is_available() must be verified before calling torch.mps.empty_cache() or the error "Cannot execute emptyCache() without MPS backend." will be raised.
This is a torch-specific behavior.

---

Public API Changes

• [*] No Public API changes
• Yes, Public API changes (Details below)

---

How Has This Been Tested?

---

Checklist

• [*] The changes have been tested locally.
• Documentation has been updated (if the public API or usage changes).
• A changelog entry has been added (see changelog/README.md), or "no changelog needed" label requested.
• [*] The code follows the project's style guidelines.
• I have considered the impact of these changes on the public API.

---

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[gemini-code-assist]

Code Review

This pull request updates the OOM error handling in src/tabpfn/architectures/tabpfn_v3.py by adding a check for MPS availability before calling torch.mps.empty_cache(). This change prevents potential errors on systems where MPS is not supported. I have no feedback to provide.

[bejaeger]

Thanks for the fix!
For the CI to succeed you will have to rebase to main and also add a chagelog. Thanks a lot.