feat: test failure triage — group batch failures by root cause

[SahilRakhaiya05]

Problem

When many tests fail in the same project (batch test run --all, regression reruns, MCP execute), agents and humans see a flat list of 20–50 separate failures. In practice, those failures often share one underlying root cause:

• Authentication / token expired
• Staging environment down (network_timeout, infra)
• Broken navigation (routing_404)
• Shared code defect (same recommendedFixTarget.reference)
• Corrupted test data / producer failure cascading to consumers

Today the CLI only supports per-test analysis (test failure get, test failure summary). An agent must download many bundles or guess which test to investigate first — wasting AI tokens, testing credits, and debug time.

Proposed solution (CLI Phase-0)

Add testsprite test failure triage --project <id>:

1. List all failed tests (GET /tests?status=failed)
2. Fetch lightweight failure/summary per test (no screenshots/video)
3. Group client-side using deterministic heuristics over existing M2.1 fields:
   • shared recommendedFixTarget.reference
   • env-wide failureKind (infra, network, network_timeout, routing_404)
   • normalized rootCauseHypothesis prefix
   • singleton fallback
4. Return clusters with representativeTestId, memberTestIds, confidence, fixPriority

Why CLI-first

• Uses only existing public APIs — no backend changes required
• Immediately reduces duplicate investigation and bundle downloads
• Natural read surface when native backend clustering ships later

Acceptance criteria

• testsprite test failure triage --project <id> --output json returns clustered output
• --type, --filter, --max-concurrency supported
• --dry-run returns canned sample
• Unit + integration tests, docs, CHANGELOG, agent skill updated

Future (backend)

Native clustering API with semantic embeddings, wave/cascade graph, and --rerun-representatives orchestration.