Fix benchmark manifest ops CI generation

[Copilot]

Description

scripts/benchmark_manifest_ops.py could not reliably run in sandboxed environments, blocking CI-generated benchmark results for the public docs site.

• Make benchmark temp-tree creation use an explicit writable --work-dir
• Add markdown/doc update modes for regenerating the manifest benchmark table
• Run the benchmark in the docs workflow before building GitHub Pages
• Replace the stale docs note with an MDX-safe generated benchmark block
• Add focused tests for work-dir usage, cleanup, and generated block replacement

Example CI invocation:

python3 scripts/benchmark_manifest_ops.py \
  --work-dir "$RUNNER_TEMP" \
  --markdown \
  --update-doc docs/src/content/docs/progress/autoloop-go-migration.mdx

Type of change

• Bug fix
• New feature
• Documentation
• Maintenance / refactor

Testing

• Tested locally
• All existing tests pass
• Added tests for new functionality (if applicable)