[AutoSP] Fix compiler correctness

[jinyouzhi]

Summary

• preserve causal-LM targets across sequence-shard boundaries and aggregate shard losses by valid-token count
• validate sequence, mesh, and attention-head layouts instead of silently dropping tokens or entering invalid collectives
• preserve non-default sequence dimensions and reconstruct supported SDPA masks after Ulysses Q/K/V exchange
• prevent stale AutoSP process-group reuse and fix FPDT GQA backward K/V gradient buffer shapes

Tests

• pytest -q tests/unit/v1/compile/test_compile_autosp.py -k AutoSPValidation
• targeted sequential AutoSP graph rewrite tests
• changed-file pre-commit hooks

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 0a392ba760

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[pengdurice]

thanks for the PR, will take a closer look later, but can you help add more details of the issues, your fixes, and the test results.

[jinyouzhi]

thanks for the PR, will take a closer look later, but can you help add more details of the issues, your fixes, and the test results.

Sure, thank you. Will add more details asap.