2.2: run scrapes from the service worker and store them in IndexedDB

[EnesYilmazcode]

Stack, merge in order: #5 > #6 > #7 > #8 (this one) > #9. Based on #7, so the diff here is only this step.
The proscan-web stack: EnesYilmazcode/proscan-web#1, EnesYilmazcode/proscan-web#2.

2.2. The content script used to own the run, so run state lived in three places and a stopped worker or a second tab could corrupt it. Now the service worker owns it (F-10, F-11, F-13, F-14, F-18).

• scripts/background/engine.js drives the run. The live record is in chrome.storage.session, and each page is written to IndexedDB in one transaction (scripts/background/db.js).
• The content script only parses and reports. The worker opens the next page with tabs.update and parses only that page; a search typed in the tab while it loads ends the run. While a page is pending, the tab sends a heartbeat, which wakes a stopped worker. That is why no alarms permission is needed.
• If session storage is emptied without an update or a restart (the extension disabled and enabled, or its process crashed), the run IndexedDB still has as live is ended as interrupted when the popup asks, and Stop ends it. The popup is no longer stuck on a run nothing can stop.
• Starting a run keeps the 10 newest runs and removes older ones, except runs still waiting to sync. When storage is full at start it drops the rest and tries once more, so the storage full message ("starting a new scan removes older saved scans") is true.
• Every message type is in scripts/lib/messages.js and goes through one router.
• Migration 3 to 4 moves 2.0 and 2.1 data into IndexedDB. It commits there before removing anything (F-26, F-101, F-102). A retry that runs after a 2.2 run no longer points latestRunId back at 2.1 or overwrites outbox rows the engine wrote.

Tested: npm test 655 jest and 35 tool tests. Playwright with one worker, before this restack: 21 passed, 1 skipped (the same F-20 sync case). That includes killing the worker over CDP between pages and swapping the extension mid-run from both 2.0 and 2.1. After the restack I ran the run tab and 503 scenarios here (all pass) and the full suite on #7 and #9. The page cap test had a race on slow runners, fixed in the last commit before this restack.

Not sure about: Chrome ignores a CDP quota override for extension origins, so storage_full is only covered in jest. I also have not checked whether a real Amazon replaceState can fire a loading event and end a run by mistake. The harness cannot show that.

Does not change permissions. The content script list loses run.js, flags.js and delta.js and gains messages.js:

[permission-lock] OK manifest.json
[permission-lock] OK dist\manifest.json
[version-gate] OK: 2.2.0 > live 2.0