LATX, fix: Reclaim guest thread and seccomp ownership resources

[LaurenIsACoder]

Scope

Rebased replacement of the draft on bfe17ae7c1. This PR now contains only guest-thread resource ownership fixes:

• Release thread-local IR2/immediate-cache allocations and non-owning TU/SMC indexes on normal exit and seccomp KILL_THREAD.
• Roll back the child CPU, private GDT, fast jump cache and filter reference on failed thread creation; preserve the parent's allocations and return the pthread error correctly.
• Reference-count immutable seccomp chains across clone, private installation, TSYNC, exit and fork. Publish an initialized child task and its inherited filter under the CPU-list lock; skip unpublished/exiting CPUs during TSYNC.
• Publish new filter roots with release ordering and acquire them in syscall readers: the exclusive guest-execution barrier does not stop a thread already handling a syscall. This post-push self-review correction is folded into the original ownership commit, with a publication-contract RED/GREEN regression.
• Release the netlink staging buffer on receive failure, drain and thread exit, restoring the caller's iovec before returning an error.

The obsolete AOT-reader changes from the draft are intentionally absent: upstream 17b479de7877 already fixed that reader, with regression coverage in 9bdce3bbe1cb. Those upstream fixes remain intact. Separate AOT merge/mapping fixes are not part of this PR.

Author self-review and combined validation

The actual published diffs were reviewed individually and checked against the local candidates. Follow-up corrections were folded into their existing commits; author/sign-off metadata updates preserved the per-PR source trees. The series remains six commits across #472, #473, #474, #475 and #476.

All five branches merge without conflicts on bfe17ae7c1. The combined source tree 80aaaaf13688387640f528d1c0bccb6a800f8d6e builds and passes 33/33 lat-pr-fast tests. On that combined binary, the 500-cycle KZT unload and retained-handle checks, 520-cycle SMC/fork checks, and 512 successful TSYNC operations alongside thread creation also complete successfully. A tests-disabled configuration registers no tests.

The integration suite still reports 27 skips, including after selecting the available guest sysroot with LATX_X86_64_SYSROOT: 25 require clang and two require the x86 cross compiler. These are not passes. This author self-review does not substitute for the upstream CI matrix, maintainer review or real-application acceptance.

Validation

LoongArch64, GCC 15.3, x86_64-linux-user, LAT O1, KZT compiled in, release (-DNDEBUG):

./configure --target-list=x86_64-linux-user --enable-latx --optimize-O1 \
  --extra-ldflags=-ldl --enable-kzt --enable-tests --disable-docs \
  --disable-werror --meson=meson
ninja -C build -j2 all meson-test-prereq
env -u LAT_LD_PREFIX meson test -C build --no-rebuild \
  --suite lat-pr-fast --print-errorlogs --num-processes 2
env -u LAT_LD_PREFIX meson test -C build --no-rebuild \
  --suite latx-integration --print-errorlogs --num-processes 2

• This isolated PR branch: 26/26 lat-pr-fast passed.
• Maintained production-body seams exercise normal/repeated/partial teardown, TU on/off, pending netlink and EAGAIN rollback, failed clone/cache initialization, TSYNC publication order, shared/private filter ownership and fork; debug and NDEBUG variants run without assertion-dependent setup.
• The original upstream implementation fails the focused resource-ownership assertions. Earlier isolated x86 thread-churn and failed-clone reproductions established the corresponding growth and cleanup paths; these are not full application acceptance.
• ASan/UBSan production-body ownership checks pass. The final source tree completes a 2,000-thread x86 churn run with bounded private-dirty memory after warm-up, 256 injected failed clones followed by a successful retry, and 512 successful TSYNC operations alongside 513 thread creations. Author/sign-off-only rewriting preserved that exact validated source tree.
• Integration suite: 27 skipped (exit 77) because the required guest-building clang/toolchain environment is unavailable. Skips are not passes.

No cache capacity changes, installed translator replacement or running-application restart. Full application memory/compatibility acceptance and unavailable integration configurations remain separate gates.