fix(doc-collector): two SPA-robustness fixes (recoverable nav error, page-state repopulation)

[gololdf1sh]

Summary

Two independent fixes that came out of running explorbot docs collect against a real-world SPA (Testomat.io beta). Each is in its own commit.

Note: Initial version of this PR included a third fix to ConfigParser.loadEnv (walk parent dirs to find .env). It regressed action-result-diff.test.ts on CI in ways the diff alone doesn't fully explain. Dropped from this PR — will resubmit separately with a CLI-level bootstrap so library semantics are unchanged.

1. fix(explorer): treat "navigating and changing the content" as recoverable

Playwright throws page.content: Unable to retrieve content because the page is navigating and changing the content on heavy SPAs whose client router rewrites the DOM mid-action (Ember, React Router, etc.). The current regex covered net::ERR_ABORTED, screenshot timeout, and font-wait — this new phrase fell through to FATAL_BROWSER_ERRORS and killed the whole crawl on the first race. Added to RECOVERABLE_NAVIGATION_ERRORS so the explorer retries instead.

2. fix(doc-collector): repopulate page state when framenavigated stripped it

After navigation, the framenavigated handler overwrites the rich ActionResult (html / links / aria) with a stripped WebPageState carrying only { url, title, statusCode }. doc-collector then reads getCurrentState() and gets state.html === undefined, state.links === []. Two consequences:

• Documentarian receives empty html → page docs degrade to a near-empty stub.
• extractNextPaths sees an empty links array → subtree crawl stops at the entry page even when many followable links exist.

Targeted fixes:

• In the main collect loop: if state.html is falsy, force capturePageState before passing to the AI documenter.
• In extractNextPaths: if state.links is empty but state.html is present, fall back to extractLinks(state.html).

Repro (combined effect)

Running explorbot docs collect /projects/{slug}/runs/{id} on Testomat.io beta:

Before	After
Crash on first action with the "navigating and changing the content" error	Crawl completes
When it didn't crash: "Pages documented: 1"	"Pages documented: 2-3" (entry + linked subpages)