[Step Time] Select primary findings by severity and iteration impact

[abhinavsriva]

Depends on:
#236
#237
#238

Populate DiagnosticIssue.score with iteration-impact fractions for typical bottlenecks and stragglers. Replace the fixed Step Time kind-priority table with severity, score, and scope tie-breaking. Preserve all qualifying secondary issues.

Minimal code surface:

• diagnostics/step_time/api.py
• diagnostics/common.py only if the shared ordering helper needs adjustment
• reporting/primary_diagnosis.py
• primary-selection and final-summary tests

Acceptance:

• A larger critical impact wins over a smaller warning impact.
• Within one severity, the larger iteration impact wins.
• A rank straggler wins only an exact impact tie with a typical bottleneck.
• diagnosis == issues[0] remains true.
• Run-level primary evidence uses the same score and denominator.