feat(pdf): separate FirePDF async cohort for crawl/batch-originated scrapes (#4406)

* feat(pdf): separate FirePDF async cohort for crawl/batch-originated scrapes

Scrapes spawned as children of a crawl or batch scrape are not
latency-bound per document — no caller is blocked waiting on one
specific PDF. Give them their own server-controlled async cohort
(FIRE_PDF_ASYNC_BULK_ORIGIN_PERCENT, default 0) so bulk PDF work can
ramp onto the FirePDF async lane independently of — and typically
ahead of — interactive traffic.

The bulk cohort hashes a prefixed key so it stays independent of the
general FIRE_PDF_ASYNC_PERCENT cohort, and all existing hard
exclusions (ZDR, short deadline, team denylist) still apply first. A
bulk scrape outside its cohort falls through to the general
percentage unchanged.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* chore: remove stray vitest log, ignore vitest.out

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>