[ci] Dashboard Grouping And Legacy V1

[Satyam-53]

Title

Update performance dashboard for v2 benchmark identity

Summary

Update the performance dashboard so v2 records are grouped by comparable identity dimensions, while keeping v1 records visible as historical context.

Scope

Dashboard grouping should include:

• workload
• variant
• benchmark version
• hardware profile
• software profile
• recipe fingerprint or recipe label where useful

V1 records should remain visible, but should not be aggregated with v2 records unless explicitly migrated.

Acceptance Criteria

• Dashboard separates variants and software cohorts.
• Dashboard does not merge v1 and v2 records in regression views.
• Historical v1 charts remain readable.
• Dashboard labels make calibration/mismatch/regression statuses clear.
• Tests cover grouping by variant and software profile.

Relevant Files

• fastvideo/tests/performance/dashboard.py
• fastvideo/performance_dashboard/
• fastvideo/tests/performance/test_dashboard_api.py
• fastvideo/tests/performance/test_dashboard_service.py
• docs/contributing/performance_benchmarks.md

Parent

RFC #1374

[SolitaryThinker]

@macthecadillac the epic endgame plan is set (owner-approved). This phase — dashboard grouping for legacy/v1 records — is yours by first refusal since it's your dashboard code: the v2 cohort grouping already shipped with the #1546 work, so what's left is the legacy/v1 grouping story. If you'd rather pass, say so and I'll pick it up. Scope as I see it: group pre-v2 records under their (model_id, gpu_type) buckets in the dashboard, visually separated from v2 cohorts, no comparator changes (those are settled — records without identity fields skip rolling comparison per the owner's policy call).