Fix Stream Creation Races (Scheduler + Metal)

[Jckwind]

Proposed changes

• Guard scheduler stream/thread vectors with a creation mutex to keep stream IDs unique under concurrency.
• Protect Metal stream_map_ with shared/unique locking and stable pointers to prevent races/rehash invalidation; centralize command buffer creation.
• Add a regression test for concurrent CPU stream creation.

Checklist

• I have read the CONTRIBUTING document
• I have run pre-commit run --all-files to format my code / installed pre-commit prior to committing changes
• I have added tests that prove my fix is effective or that my feature works
• I have updated the necessary documentation (if needed)

[zcbenz]

MLX does not support multi-thread use cases, we don't oppose making changes for thread safety if it overall improves code quality but I don't think this can be categorized as such.

[Jckwind]

cool with me - I’ve made my fork of MLX thread and stream safe.