fix(verdict): render per-case stdout in Run modal (reverses BRAT #2)

[LikeSundayLikeRain]

Summary

The Run verdict modal showed the same combined stdout under every test-case tab. Each tab should show only its own case's print() output.

Root cause

The BRAT #2 fix (2026-06-05) at src/solve/verdictModalRenderer.ts treated code_output as a flat run-global blob. The reasoning was: when LC returns code_output as string[], each element looks like one line of combined stdout — verified against node_modules/@leetnotion/leetcode-api/lib/index.js:1887, where formatTestOutput joins the array on \n.

The leetnotion library does join on \n, but the wire shape is element-per-CASE, not per-line. The library normalization is what's flat; the underlying response isn't.

Verification

Live probe against leetcode.com/problems/two-sum/interpret_solution/ on 2026-06-11 with a 3-case Python3 solution that prints f"CASE-MARKER nums={nums} target={target}" per case. Response:

total_testcases: 3
code_output: [
  "CASE-MARKER nums=[2, 7, 11, 15] target=9",
  "CASE-MARKER nums=[3, 2, 4] target=6",
  "CASE-MARKER nums=[3, 3] target=6"
]
std_output_list: [
  "CASE-MARKER nums=[2, 7, 11, 15] target=9\n",
  "CASE-MARKER nums=[3, 2, 4] target=6\n",
  "CASE-MARKER nums=[3, 3] target=6\n",
  ""
]

Both fields carry per-case stdout. std_output_list is the modern canonical name; code_output is a deprecated alias (the xuhuafeifei/leetcode-runner Java model marks code_output @Deprecated with the comment "this field is not used anymore, please use std_output_list").

Changes

• src/solve/types.ts — type std_output_list?: string[] and expected_std_output_list?: string[] on RunCheckResponse (previously fell through the [k: string]: unknown index signature).
• src/solve/verdictModalRenderer.ts — renderActiveCase reads per-case stdout via splitOutput(res.std_output_list ?? res.code_output, arity)[activeIdx] with the trailing \n stripped. The code_output fallback covers older or partial response shapes (e.g., the existing run-sample.json fixture has empty code_output: [] despite carrying std_output_list).
• tests/solve/verdictModalRenderer.test.ts — replaces the BRAT feat: v1.1 — Contest, AI Coach, and Preview #2 line-per-element regression tests with per-case wire-shape tests using the live probe payload. New assertions:
  • Each tab shows only its own case's stdout (clicking through Case 1/2/3 swaps the marker).
  • Trailing \n from std_output_list is stripped for display.
  • Falls back to code_output when std_output_list is absent.
  • Empty padded slots (e.g., LC pads to 4 entries for 3 cases) render no Stdout section.

Test plan

• npx vitest run tests/solve/verdictModalRenderer.test.ts — 32/32 pass (4 new + 28 existing).
• npm run build (tsc --noEmit + esbuild prod) — green.
• npm run lint — green (1 unrelated unused-import warning).
• Live verified the wire shape with a real LC API probe (Two Sum, Python3, 3 cases).
• Deployed to interview-prep BRAT vault for manual verification.

The full vitest run shows 5 timeouts in unrelated AI/preview test files (tests/ai/clearKey.test.ts, tests/ai/probe-debounce.test.ts, tests/ai/reset-disclosures-command.test.ts, tests/preview/router.test.ts). These are pre-existing flakes under load — running them in isolation off main (without my changes) reproduces the same flake-vs-pass behavior. Confirmed by stashing my diff, re-running just those files: all 38 tests pass in 3.45s.