crawl-all: enqueue full crawls via Pueue instead of running inline

[MirjamOdile]

Problem

Since production crawls auto-detach into Pueue, a single scrapai crawl is
parallel-safe and disconnect-safe — but crawl-all deliberately bypassed that
(detached=True forced every spider inline). So "run the whole project" meant:
strictly sequential, one spider at a time, bound to the terminal, and the whole
batch dies on Ctrl+C or an SSH drop. A 15-spider project would block a terminal
for hours with no resumability.

Change

crawl_all now calls each spider through the SAME path as a single crawl
(detached=False): a full run submits itself to Pueue and returns, so the
whole project is queued in seconds, runs at whatever concurrency the Pueue
group is set to, and survives disconnects. --limit N keeps the old inline
sequential behaviour (test crawls are foreground by design). A closing hint
points at ./scrapai crawl-status / pueue status.

Verification

• test_crawl_all_enqueues_each_spider — every spider goes through the detach
  path (detached=False), the queue hint prints.
• test_crawl_all_with_limit_stays_inline — --limit runs stay inline, no
  queue hint.
• Full unit suite green; used live to queue a 15-spider project.

Trade-offs

• Pueue becomes required for crawl-all full runs (it already was for any
  single full crawl; the error message is the same).
• Sequential-inline behaviour, if ever wanted, is one pueue parallel 1 away.
• If a pueue add fails mid-loop, spiders after the failure aren't queued
  (the CLI exits with the same clear error a single crawl gives).