fix(lingbot-vla): align service validation and replay gates

[HappyDog060713]

Restore strict validation for the standalone generic VLA server while retaining the explicit skip-validation development override. Apply the same numerical replay thresholds to BF16 and quantized release profiles, keeping exact equality as diagnostic metadata.

Update tests and documentation for the validated H100 profiles and service behavior.

Verification: 170 relevant tests passed; ruff and diff checks passed; strict pipeline validation passed; BF16 WebSocket-to-MuJoCo smoke completed; five in-scope release profiles passed with AIPerf.

Description

Motivation

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Performance improvement
• Code refactoring
• Documentation update
• Other (please describe):

Changes Made

Testing

• Unit tests pass (pytest tests/)
• Manual testing performed
• Benchmarks added/updated (if applicable)

Test commands:

# Add the commands used to test your changes
pytest tests/

Checklist

• Code follows the project's coding standards (ruff)
• Pre-commit hooks pass (pre-commit run --all-files)
• All tests pass (pytest tests/)
• New tests added for new functionality
• Documentation updated (README, CLAUDE.md, docstrings)
• Commit messages are clear and descriptive
• PR title follows the convention: [TYPE] Brief description

Related Issues

Fixes #

Additional Notes

GPU Architecture Support

• SM80 (Ampere, Ada Lovelace)
• SM90 (Hopper H100)
• SM100+ (Blackwell)

Performance Impact