test(web-e2e): say WHICH subsystem failed when the deep route never serves

[choraria]

The CI failure on 2026-08-01 reported last probe: 404 (no Location) for the full 180s — the catch-all answered every poll, so the deepest route never served. The message then blamed the scan, which it cannot know.

"Never served" has two unrelated causes:

• the scan never reached the route (next dev still walking src/app), or
• the route was registered and never compiled/served — a different subsystem, scan exonerated.

Reading the old message sends you through src/app discovery even when next dev's own route table already listed the route.

Change

awaitRouteScan returns which of three things it saw, and the failure quotes it:

outcome	what it means
listed	registered and still 404 — not the scan; look at compile/serve
missing-route	the table was being written and this route never appeared — the scan really didn't reach it
absent	no readable types file, so the gate was inert and proved nothing either way

That last one is stated explicitly because "the scan was fine" is the one conclusion an inert gate can't support.

⚠️ Scope

This does not fix the flake and isn't meant to. It turns one ambiguous timeout into a message that names a subsystem, so the next occurrence is evidence rather than another guess.

Test plan

• pnpm lint — 0 fail; real suite 25 passed in 31s
• 4 unit tests (injected read/sleep) covering all three outcomes plus early return
• Forced a real timeout: harness printed route scan: listed — … so this is not the scan, correct locally

🤖 Generated with Claude Code