x86: clonevm_exec_test is saved blocked in a kernel wait and never restored, on pristine main — the unnamed half of the 'USERSPACE TEST COMPLETE absent' stall

[ryanbreen]

Summary

On beast breenix-x86 (KVM, -cpu host, testing,external_test_bins),
clonevm_exec_test's thread is saved blocked in a kernel wait and never
restored, on pristine main as well as on branches. The boot does not die —
other processes keep starting, blocking, waking and exiting for hundreds of
lines — but init is left waiting on it, so USERSPACE TEST COMPLETE never
appears and the gate reports only "boot did not finish".

Named by scripts/x86-strand-census.sh on a pristine-main serial preserved
in-repo:

strand census: thread 23 (clonevm_exec_test) saved blocked at line 3594 and never restored (339 further lines followed)
STRAND_CENSUS: threads_saved_blocked=8 stranded=1 lines=3933

Why it is being filed now

This shape has been showing up for two rounds as the unattributed
USERSPACE TEST COMPLETE was absent red, credited to "main" without a name.
It has a name now: until x86-strand-census.sh landed, the gate had no check
that could see a thread which is silenced rather than failing, so the first
cause was invisible and only the terminal symptom was reported. Two different
first causes — this one and the poll_tcp_oracle strand that #568's reorder
fixes — produced an identical gate message for a whole round.

Rate

• pristine main @ 6d0d2ac5, beast KVM: 3/8 boots show the
  USERSPACE TEST COMPLETE absent stall, and the preserved specimen strands
  clonevm_exec_test.
• round-3 A/B, reorder-reverted arm: 2/16 boots strand clonevm_exec_test.
• round-3 A/B, merge-candidate arm: 0/24. The candidate arm is not a fix
  for this — 0/24 against 2/16 does not discriminate at this sample size — it
  simply did not occur there.

Serials (in-repo)

• docs/planning/green-program/sockets/serials/x86-boot-did-not-finish-MAIN-kernel-20260829.txt
  (pristine main; strands clonevm_exec_test only)
• docs/planning/green-program/sockets/serials/x86-boot-did-not-finish-armA-kernel-20260829.txt
  (strands poll_tcp_oracle first at line 2574, then clonevm_exec_test ~1000
  lines later — the two causes side by side)

Relationship to #690 and #610

#690 is the aarch64 clonevm_exec_test post-exec rendezvous stall (1/30,
cortex-a72) and #610 is the racy sibling assertion in the same test. This is the
x86 shape and is filed separately because what is observed here is a
never-restored kernel wait named by the context-switch record, not a rendezvous
that completes late. They may share a cause; nothing here establishes that.