[fglock]

Continue the explicit owner-slot migration across #1649, #1618, and #1860.

This draft carries the no-sweeps ownership work, including exact owner identities, caller argument cell aliasing, weakly tracked cleanup that respects surviving owner slots, and interpreter pad-constant ownership. It is related to draft #1816, which carries the earlier Mojolicious/Catalyst controller and WebSocket lifetime work.

The interpreter materialization fix transfers pad constants from the temporary executable wrapper to the published Perl-owned CV. The focused code-pad final-slot regression seeds a positive legacy count to verify that owner identities can release the final slot; this is an invariant test and does not claim to reproduce natural count drift. The separate nested untracked-owner test has a real counted scalar owner (refCount == 1) and verifies that final-cell cleanup retires the nested edge without disturbing that independent owner.

The cleared-aggregate scope-exit path now checks direct Perl owner identities before using the cached external-root compatibility query. The fallback remains for ownerless cases that have not yet migrated. Queued owner releases now remain visible through their deferred count transition, so scope-exit finalization cannot race a release already in flight.

Validation on this update:

• nice -n 19 make — passed in 7m38s; includes the package-root, cleared-aggregate, and queued owner-release regressions.
• ClearedAggregateOwnerIdentityFinalizationTest — passed; live package-owner identity avoided external-root queries and snapshots.
• Pad-constant ownership test — system Perl 10/10; JVM and interpreter 10/10 each with corrective sweeps disabled.
• HTML-Tree 5.07 upstream suite — all 23 files passed: system Perl 969 tests (the leak test skipped because Test::LeakTrace is unavailable); JVM and interpreter 989 TAP checks each with corrective sweeps disabled after the cleared-aggregate change. PerlOnJava’s 20 leak-test checks use its no-op Test::LeakTrace stub and are not leak measurements. t/refloop.t passes 8/8 on all three runtimes.
• Lifecycle matrix — system Perl, JVM, and interpreter each passed 104 assertions; PerlOnJava backends ran with corrective sweeps disabled.
• Post-change weak/DESTROY/resurrection/cycle matrix — 8 fixtures, 68 assertions on each backend with corrective sweeps disabled.
• Mojolicious controller and WebSocket lifetimes regress with #1642 integration #1794 controller and WebSocket cases — system Perl 15/15; after this change, JVM and interpreter each passed the controller fixture 3/3 and WebSocket fixture 6/6 with corrective sweeps disabled.
• Avoid repeated live-root scans over growing lexical aggregates #1860 growing lexical output — system Perl and both PerlOnJava backends 4/4; after this change, JVM and interpreter each passed the fixture 4/4. Data::Tumbler stress exercised 65,536 results on all three.
• Added issue_1649_package_nested_untracked_weak_owner.t; system Perl, JVM, and interpreter each pass 7/7 with corrective sweeps disabled.
• lychee --offline dev/design/perl-ownership-rewrite.md — passed (14 links).
• Cleared-aggregate legacy-count invariant — the Java test assigns a positive count directly; it checks the owner-identity boundary but is not evidence of natural count drift.
• Pending owner-release finalizer regression — JUnit passed; a real queued counted-owner release keeps the weak observer live until the deferred decrement drains.
• Async weak/DESTROY reproducer — JVM and interpreter each pass 2/2; the equivalent standard-Perl weak-reference/DESTROY oracle passes 2/2.
• nice -n 19 make check-links — passed (791 links checked, 0 errors).

The ownership rewrite remains in progress. Remaining work includes the package-root compatibility pin, remaining reachability and sweep fallbacks, the rest of the ownership families, then rerun the HTML-Tree suite after those changes. Keep this PR in draft while those paths are migrated and validated.

This snapshot starts at 4d5e0841e1147231314cfec0752096fd1f381d75; GitHub master has advanced since then. Upstream integration and merge-conflict review remain before this draft can be considered merge-ready.