[Epic][TypeSpec Authoring][Benchmark] Continuous improvement

Epic to track continuous improvement work for the TypeSpec authoring benchmark — ongoing investments after the initial Vally adoption (#14667) is done.

Scope (non-exhaustive, will grow over time):

- Continuously add and refine benchmark cases (basic & advanced dataplane, basic & advanced ARM, regression cases for real-world issues).
- Improve grader quality and judge prompts to better reflect TypeSpec authoring correctness.
- Reduce flakiness and total running time of the benchmark suite.
- Strengthen reporting and dashboards (per-run timing, pass-rate trends, per-category breakdowns).
- Track and adopt upstream Vally features/fixes that improve our benchmark experience.
- Tune skill prompts / instructions based on benchmark findings.