[meta] PR Review Companion improvements

What would you like to see added to BCD?

Extend PR Review Companion to bundle reusable PR analysis steps, combining their results into its existing comment and automatically managed labels.

A pragmatic starting point:

• Run analyzers against the same comparison commits and aggregate their results centrally.
• Preserve the existing diff links and add concise, actionable findings to the same comment.
• Refresh findings and synchronize managed labels as the PR changes, preserving unrelated labels.
• Distinguish unavailable analysis from an analysis with no findings, and prevent outdated runs from overwriting newer results.
• Keep analysis runnable locally, with CI running trusted analysis code against PR data.

Initially, the Companion can orchestrate existing scripts. Before refactoring diff:flat or extracting shared comparison logic, add regression coverage for its existing behavior, including grouping, mirroring, and added or removed features. Its current test coverage is limited.

Initial sub-issues:

• #30261: Detect newly partial implementations and synchronize the "sets partial" label, reusing the work in #30385.
• #30483: Highlight unchanged keys when most keys belonging to a web-features entry change.
• #28640: Summarize changes by web-features entry, potentially broken down by browser. Coordinate this with #30483.

How impactful do you think this enhancement will be?

Help reviewers spot potentially missed updates and find PRs that need discussion, while keeping feedback in one place. Shared orchestration and publishing should make additional checks easier to add and maintain.

Do you have anything more you want to share?

#30385 provides an initial implementation for partial-implementation detection and labeling. This meta issue tracks integrating that capability into the broader Companion workflow.

The web-features findings should be advisory. Thresholds and browser-specific comparisons can be refined in the linked issues.

Start by reading the existing PR Review Companion scripts and the diff:flat behavior, then inspect regression coverage for grouping, mirroring, and added or removed features. Use #30261, #30483, #28640, and #30385 to map the initial analyzers and coordination boundaries. Done means analyzers share comparison commits, findings and managed labels refresh safely, and unavailable analysis is distinct from no findings.