docs: refine code-review skill with conformance, cache, and async checks

[allenporter]

Summary

Builds on #988 and #953 to sharpen the code-review agent skill based on recent PR review themes (#955, #966, #971–#974):

• Automated Conformance vs. Semantic Focus: Notes that structural invariants (RoborockBase, RoborockEnum, RoborockException, trait boundaries) are enforced in tests/conformance/, instructing reviewers to flag unjustified allowlist additions while focusing deeper review energy on semantic, lifecycle, and design issues.
• Event Loop & Cache Invalidation Checks (Layer 2): Adds explicit checks for keeping CPU-intensive work (map decompression, PIL rendering) off the asyncio event loop and ensuring derived caches (e.g., base maps, calibration) are properly invalidated and never mutate cached telemetry state.
• PR Scoping: Adds a check for PRs that mix unrelated subsystems or combine large protocol/parser additions with trait consumption.
• Output Formatting: Tailors output guidance for both interactive summary reviews and concise inline PR bot comments while avoiding style nitpicks already covered by Ruff/Mypy.

Related to #953

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.