Pre-existing test failures in dropped-language-gap.test.ts on main

[carlos-alm]

Both tests in tests/integration/dropped-language-gap.test.ts (added in #1123) currently fail on main (commit c7e600a):

• Dropped-language gap repair on quiet incremental (#1083) — earlyExit=true > repairs the missing file node when the orchestrator returns earlyExit=true
• Dropped-language gap repair on dirty incremental (#1083) — earlyExit=false > repairs the missing file node alongside a real change (earlyExit=false)

Both fail with expected undefined to be defined on the readFileNodeRow lookup — the backfill apparently isn't writing the expected row when these tests exercise it.

Surfaced while resolving #1144. Filing separately to keep that PR focused on the dead-code removal.

[carlos-alm]

Investigated — could not reproduce on main (a029d43) or at the referenced c7e600a:

• Local run on origin/main: both tests pass.
• Local run at c7e600a (temporary worktree, same node_modules + grammars): both tests pass.
• CI history: PR fix(native): backfill new dropped-language files on quiet incrementals #1123 (where the tests were added) green on Linux/macOS/Windows; c7e600a's main CI green; latest completed main run (81150e4) green.
• The only pipeline.ts diff between c7e600a and origin/main is fix(builder): remove duplicate early-return in backfillNativeDroppedFiles #1148's removal of a duplicate if (missingAbs.length === 0) return; — semantically identical, no behaviour delta.

Best guess for the original report is a stale local environment (out-of-date native binary, partially-built node_modules, or test pollution from neighbouring files in a npm test run).

Rather than close on "couldn't reproduce", #1169 adds an isolated CI step that runs only dropped-language-gap.test.ts on every PR across all three OS runners. If the gap-repair path ever regresses with the exact bug shape this issue describes ("node row never re-inserted"), the canary will fire under its own clear name — and because it runs in isolation, it also rules out test pollution as a cause.

Leaving this issue open for you to close once #1169 merges and you confirm the canary is doing its job.