feat(experimental): integrate TPU-Raiden weight synchronization

[khatwanimohit]

• Implement transport-neutral weight synchronization subsystem in tunix/experimental/orchestrator (WeightSyncCoordinator, WeightSyncHandler, WeightSyncSource, WeightSyncDestination, WorkerRoundTracker).
• Add fail-safe 3-phase quiesce-and-commit barrier (prepare -> transfer/h2d -> commit) with prefix cache invalidation and rollback on failure.
• Implement dual backend support:
  • local_launcher: uses native TPU-Raiden WeightSynchronizer API with dynamic host IP resolution.
  • pathways: uses weight_synchronizer_ffi (init_weight_synchronizer, multi_h2d) for Shared Pathways Service (SPS) and GKE multi-slice execution.
• Ensure deterministic parameter manifest indexing via alphabetical ordering across PeftTrainerV2 and LegacyVllmSamplerAdapter.
• Upgrade dependencies to JAX 0.11.0 and Flax 0.12.8 with native runtime support (zero compute_on monkey-patching).
• Update GKE Kubernetes JobSet launcher (k8s_launcher.sh) and runner node entrypoints for distributed multi-slice deployments.
• Add comprehensive unit and integration tests in tunix/tests/experimental_weight_sync_test.py.

Resolves #<issue_number_goes_here>

It's a good idea to open an issue first for discussion.

Reference

Colab Notebook

Checklist

• I have added all the necessary unit tests for my change.
• I have verified that my change does not break existing code and all unit tests pass.
• I have added all appropriate doc-strings/documentation.
• My PR is based on the latest changes of the main branch (if unsure, rebase the code).
• I have signed the Contributor License Agreement.
• I have followed Contribution Guidelines.