Add e2e tests for run-detail depth

[alchaplinsky]

Summary

• Adds tests/e2e/specs/runs/detail.spec.ts covering the gaps left by runs/view.spec.ts.
• New tests (5):
  • Failed sidebar filter shows only failed runs
  • Failed scenario run renders the "Run failed" banner with the parsed error_json.message
  • Multi-step scenario trace renders Prompt Assembled → Model Requested Tool Call → Tool Call → Tool Response → Final Model Response in order
  • Expanding a model step reveals the USAGE row (Prompt / Completion / Total)
  • Expand all / Collapse all in the metrics bar toggle every step body

Test plan

• bun run test:e2e -- --spec ./tests/e2e/specs/runs/detail.spec.ts → 5 passing, ran 3× consecutively to confirm stability

🤖 Generated with Claude Code

[greptile-apps]

Greptile Summary

Adds tests/e2e/specs/runs/detail.spec.ts with five new end-to-end tests covering run-detail behaviour that was previously untested: failed-run filtering, the error banner, multi-step tool-call trace ordering, token usage display, and the Expand all / Collapse all control.

• Seeds three scenario executions (multi-step with tool calls, failing with error_json, and a simple succeeded run) in beforeEach; relies on the existing global afterTest hook for database cleanup between tests.
• Tool-step selectors use the → / ← arrow characters that are unique to Tool Call and Tool Response labels, avoiding the ambiguity flag raised in the earlier review thread.
• The "Expand all / Collapse all" assertion counts all <pre> elements page-wide and expects exactly 5; this is slightly fragile if the page chrome ever gains additional <pre> nodes, but matches the pattern already established in view.spec.ts.

Confidence Score: 5/5

Additive test-only change with no production code modifications; safe to merge.

The change is purely additive (one new spec file, no production code touched). The selector ambiguity flagged in the earlier review thread has been addressed. Database isolation between tests is handled by the existing global afterTest hook. No defects were found in the test logic.

No files require special attention.

_{Reviews (2): Last reviewed commit: "Tighten tool_call selectors to use the a..." | Re-trigger Greptile}