perf: replace chunk-based pipeline with streaming worker pool

Removes sequential chunk barriers (30 PRs per batch). Each PR now flows
independently through enrich -> filter -> pre-classify -> deep process
using FuturesUnordered with per-stage semaphores (5/15/5 concurrency).
AtomicUsize counter stops the pool as soon as max_tasks is reached.

Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>

Author: echobt