devboxverify full run blocks its last journey: the 480 s journey budget no longer covers 50 journeys #439

Problem

node devboxverify/main.js --epic 424 on main 4bb2c32 ran 49 of 50 journeys and reported JOURNEY add-browser-in-window BLOCKED run budget spent, so it posted error for devbox/verify-full. The journeys' timings summed to 480,190 ms against JOURNEY_BUDGET_MS = 480000 (devboxverify/main.js:17). The two journeys eve#432 added (cos-outside-simple, about 59 s, and cos-outside-hooks) pushed a full run past the budget, which has not changed since #96. No full or nightly run can pass now.

Root cause

The shared journey budget was sized for fewer journeys (design A12: "keeps eve's nightly under 10 minutes"). It is a run-time cap, not a check: when it runs out, the remaining journeys are BLOCKED.

Fix

Raise JOURNEY_BUDGET_MS to 600 s and update README and design A12 to match (the nightly stays under about 12 minutes, well inside its 45-minute kill). No journey, assertion or timeout changes. This narrows the BLOCKED path; it widens nothing.

ship-blocker: no

Found while working #424.

Contract

Ownership: devboxverify/main.js (the JOURNEY_BUDGET_MS constant only), devboxverify/README.md (the budget sentence), docs/design-devboxverify.md (A12 only).
Proof: red on main is the 4bb2c32 full run above. Head harness full run (no post) runs 50 of 50 with no budget BLOCK. Break-and-restore: with the budget cut to a value smaller than one journey, an --only run reports BLOCKED run budget spent; restored, it runs.
Waits: run exit and its SUMMARY line.

Decisions made for the owner

1. 600 s, not more: it fits today's 50 journeys with about two minutes to spare and keeps the nightly near 12 minutes.
2. Amendment (doc only): docs/FEATURES.md (the Notes' nightly budget line), so the figure matches; the reviewer found it stale.
3. The head full runs were cut short by the local Chat model slowing down at 13:20 (median request 1 s to 8-14 s, up to 69 s), not by the change. The proof that a full run fits is the --epic 424 run on main after this merges, once the model is back to normal latency.