[PiUnknown]

Fixes #2380

Problem

When a streamed deep crawl is abandoned early (caller breaks out of async for without closing the generator), deep_crawl_active stays True in the caller's context. The flag was only reset in the generator's finally, which runs on close/GC - too late, and possibly in a different context. The next arun() with a deep-crawl config then saw the flag still set, treated itself as a nested call, and fell through to a single-page crawl.

Fix

Manage the flag around each yield of the stream wrapper instead of relying on generator cleanup:

• the flag is True only while the wrapper is actively advancing the inner generator (when nested arun calls actually happen)
• it is reset while suspended at the yield, so an early break leaves the caller's context clean and the next crawl starts fresh
• a guarded reset(token) (falls back to set(False)) handles consumption from a different context/task
• the wrapper's finally also aclose()s the inner generator so strategy cleanup isn't left to GC
• non-stream path now uses a proper reset(token) and holds the flag across the strategy call itself

Testing

• New end-to-end regression test: local aiohttp server, full crawl (6 pages) -> stream stopped early at 2 without closing -> next crawl returns all 6 (previously returned 1).
• Issue repro before fix: full 6 / stopped 2 / next 1. After fix: full 6 / stopped 2 / next 6.
• pytest tests/deep_crawling/: 87 passed, 1 pre-existing failure in test_deep_crawl_perf_2242.py (wall-clock timing flake on Windows, unrelated).