Emulator: socket churn under packet-pool pressure times out (no DH0:.done) on every run

[tinic]

The Emulator tier's "Socket churn under packet-pool pressure" arm (endurance harness, loopback mode, 60 s, 4 connections, probes=1) times out on every recent run: the guest never writes DH0:.done.

Run	SHA	Result
36119295584	main 1d60dad	TIMEOUT after 360s (no DH0:.done), status=fail rc=124
36174506993	fix/emu-dhcp-churn a99edcc	same
36204164690	main 58c5c4f (release candidate)	same; runner build/amiberry-testhd-ci-endurance has no .done, serial log holds only the ANXD-RUN token

Earlier in that 36174506993 run, the guest printed repeated recv rc=-1 errno=35 blocking=0 lines (EWOULDBLOCK on non-blocking reads) before the timeout.

It predates the 2026-09-25 merge train and is independent of the UDP 5353 loopback defect. Next step: capture the guest state at the hang (stack first, since the Amiga has no MMU), not another rerun.

🤖 Generated with Claude Code

[tinic]

Fixed on main (merge f2649e2 of fix/67-adopt-evict 477d674). Root cause: the adopt pool has 16 slots (1 reserved) and each bsdsocket base's cached adoption held one for life, so the churn test's 16th+ opener parked forever in NetStackQuery. Fix: a full pool evicts a DORMANT|CACHED adoption; the evictor clears signal_owner under Forbid before discard, and the owner frees its run signal exactly once on re-adopt/release. Evidence: churn test pre-fix TIMEOUT 360s x2, post-fix RC=0 61s 824 xacts, 0 corrupt/desync. adopt_pool tests cover the same-Task negative probe; host 468/468. Sizes: default 363,896/364,000, minimal 237,468/238,000, micro 203,368/204,000.