Add validation to block unsupported parameters in post-training SFT, …

[shralex]

…RL, and DPO

• SFT, RL, and DPO trainers now error out on optimizer_memory_host_offload=True since the underlying Tunix trainers do not support host offloading.
• DPO and RL trainers now error out on num_vocab_tiling > 1 since they require full vocabulary projections to compute preference/policy log-probabilities, which is incompatible with vocabulary tiling.
• Created unit tests for SFT, RL, and DPO validation functions to verify these restrictions.

FIXES: b/520780393

Tests

Added unit tests

Checklist

Before submitting this PR, please make sure (put X in square brackets):

• I have performed a self-review of my code. For an optional AI review, add the gemini-review label.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run end-to-end tests tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed, including adding new documentation pages to the relevant Table of Contents (toctree directive) as explained in our documentation.

[codecov]

Codecov Report

❌ Patch coverage is 81.25000% with 3 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/maxtext/trainers/post_train/dpo/train_dpo.py	83.33%	1 Missing ⚠️
src/maxtext/trainers/post_train/rl/train_rl.py	83.33%	0 Missing and 1 partial ⚠️
src/maxtext/trainers/post_train/sft/train_sft.py	75.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!