Add an in-page fetch() path so backend code can run a same-origin fetch from
inside the warm camoufox page (real fingerprint + context cookies + per-provider
proxy) and get back {status, body}. Purely additive: /fetch, _fetch_once,
_do_fetch, fetch_handler and BrowserFetcher.fetch are untouched. Nothing in the
prod flow calls it yet — a later PR wires avito IMV to it.
Sidecar: _fetch_json_once navigates to origin (anchor page, short 500ms settle
instead of full BROWSER_WAIT_MS), runs fetch via page.evaluate, mirrors the
_fetch_once page lifecycle + recycle/relaunch. _do_fetch_json mirrors _do_fetch
crash-retry. fetch_json_handler mirrors fetch_handler (400/503/500 handling).
Client: BrowserFetcher.fetch_json posts to /fetch-json with the same
single-retry-on-HTTPError/TransportError pattern as fetch.
Tests: browser/test_server_fetch_json.py (handler happy path, origin navigation,
POST payload passthrough, 400 missing url, 400 bad json, 503 unavailable, 500 on
evaluate error) and backend fetch_json client tests (payload, defaults, retry,
raise-after-two-errors).
Refs #915
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
