Tighten reportability checks and workload scoping in study comparison

[lullu57]

Summary

Run-level reportability and study-level filtering are too loose. Study comparison does not enforce run-group workload scoping, and summary.json reportability is weaker than the documented contract while still being used as the admission gate for study aggregation.

Evidence

• src/kernel_tuner/analysis/reporting.py:392
• src/kernel_tuner/analysis/reporting.py:403
• src/kernel_tuner/analysis/comparison.py:232
• src/kernel_tuner/analysis/comparison.py:291
• configs/studies/validation_phase.yaml:120
• docs/specs/analysis_and_reporting.md:79

Why This Matters

Study-level evidence can silently mix workload classes or trust a run summary that did not fully re-check reportability requirements. That weakens paper-facing conclusions.

Expected Fix

• Enforce run-group workload scoping in study comparison.
• Tighten run-level reportability checks to match the documented contract.
• Make study admission depend on the stronger contract, not a loose boolean.

Acceptance Criteria

• Study run-groups respect configured workload scoping.
• is_reportable reflects the documented policy, not a reduced subset of checks.
• Tests cover study filtering and reportability enforcement.