[guyoron1]

Summary

Fixes #2420

When rtk test <cmd> doesn't recognise the runner (make test, bazel test, a custom script), it fell back to OUTPUT (last 5 lines):, which hid the failure whenever it wasn't in the tail. #2420's script prints FAIL: test_login broke among 50 PASS: lines, and only the last five PASS: lines came out.

Now, only when nothing was recognised (no summary, no runner diagnostic block) and the exit code is non-zero:

• A. the error blocks rtk err uses (ErrorStreamFilter, reused rather than copied), plus a bare fail / fails word marker that only test runs read (ErrorStreamFilter::for_test_run(); the shared ERROR_PATTERNS and rtk err are untouched) -> [FAIL] DETECTED FAILURES: capped at CAP_LIST with ... +N more
• B. otherwise [FAIL] Command failed (exit code: N) plus the last ten non-blank lines (same text as rtk err)

A run that exits 0 keeps today's tail, and recognised runners that print a summary are byte-identical (golden tests). The tee recovery hint on failures comes from the existing print_with_hint, and never_worse still applies. A recognised runner that died before its summary (for example a cargo test compile error, exit 101) now shows the error[E0425] block instead of the last 5 lines.

Changes from the first version of this PR, found by re-running #2420's exact script:

• it did not fix the issue's own repro, because FAIL: isn't in ERROR_PATTERNS; hence the bare-fail marker
• it ran the new path regardless of exit code, so a green run printing warning: and Summary: 0 failed, 30 passed got a false [FAIL] DETECTED FAILURES: banner
• the third tier ([ok] All tests passed (no recognized test runner)) is dropped: it claims something rtk doesn't know (cargo test --no-run and pytest --collect-only land there too) and it discards the tail that carries things like 12 tests collected. Restoring it is a three-line change if you prefer it.

The code has moved to src/core/runner.rs since the first version (run_test_cmd / extract_test_summary), so it is re-applied there.

Before/after on #2420's script (rtk test ./vt.sh, exit 1):

develop : OUTPUT (last 5 lines): PASS: case_46 ... PASS: case_50
this PR : [FAIL] DETECTED FAILURES:
            FAIL: test_login broke

Tests

• 9 unit tests, including the golden table with an exit-code column (expected outputs untouched) and the bare-fail word boundaries (test_fail_case does not match)
• tests/generic_test_runner_test.rs (unix, 3 e2e tests): rtk test shows only the last 5 output lines (not failures) for unrecognized runners #2420's script, a green run is never reported as failed, and error blocks
• all new tests fail against the old fallback
• cargo fmt --all -- --check, cargo clippy --all-targets (0 warnings), cargo test --all (all pass)

[guyoron1]

Rebased onto current develop. The code moved to src/core/runner.rs (run_test_cmd / extract_test_summary), so I re-applied it there. Re-testing with the exact script from #2420 showed my first version still hid FAIL: test_login broke (ERROR_PATTERNS has no bare FAIL), and it printed a [FAIL] banner for green runs that merely print "warning" or "0 failed".

Now only a non-zero exit takes the new path: the rtk err error blocks plus a bare-fail marker that only test runs read (the shared ERROR_PATTERNS are untouched), otherwise the exit code plus the last ten lines. A run that exits 0 keeps its tail, and recognised runners are unchanged. Unit and e2e tests fail against the old fallback.

Two notes for review: I dropped my third tier ([ok] All tests passed), since it claims something rtk doesn't know and discards the tail that carries things like 12 tests collected; restoring it is a three-line change if you prefer it. And #4125 also edits src/core/runner.rs, so expect a rebase for whichever lands second.