Add PR review performance telemetry and dashboards

[dayland]

What & why

Extends the BC PR Review category added in #774 with performance telemetry, experiment provenance, and dashboard visibility. This is enablement only: it does not add or publish a new experiment result set.

Changes

• Collect Copilot token usage, API-call count, estimated AI credits, and elapsed time from the engine transcript or structured run metrics.
• Record BCQuality knowledge articles retained and pruned by the review filter.
• Aggregate performance fields only for Code Review, preserving unrelated category result and leaderboard schemas.
• Add Quality and Performance tabs to the Code Review experiment leaderboard, including engine/BCQuality revisions and self-contained labels for legacy runs.
• Make BC-ALAgents source selection symmetric with BCQuality through repo, ref, and local_path configuration plus matching CLI overrides.
• Resolve and record exact BC-ALAgents and BCQuality commits for reproducibility.
• Let BC-Bench perform authenticated source checkout instead of hardcoding BC-ALAgents main in the workflow.

Experiment configuration

A private BC-Bench branch can independently select either source under pr_review in src/bcbench/agent/shared/config.yaml:

pr_review:
  engine:
    repo: microsoft/BC-ALAgents
    ref: main
    local_path: null
  bcquality:
    repo: null
    ref: null
    local_path: null

Branch names, tags, and commit SHAs are supported. Local checkout overrides are available for unpushed development.

Validation

• Ruff format and lint
• Changed-file type checks
• 762 tests passed, 2 skipped
• Jekyll dashboard build

AB#645219