Fix Sub::Quote weak reference cleanup

[fglock]

Fix Sub::Quote leaks.t tests by clearing weak refs to CODE objects

Problem

The Sub::Quote leaks.t tests were failing because weak references to CODE objects were never cleared, preventing Sub::Quote::CLONE from cleaning up expired entries.

Solution

Modified WeakRefRegistry.clearWeakRefsTo() to always clear weak refs to CODE objects by passing true instead of false to the includeCode parameter. This allows Sub::Quote::CLONE to properly clean up expired entries.

Test Results

• leaks.t: All 9 tests now pass (previously failed 4 tests)
• quotify.t: All 2595 tests pass
• hints.t: 2 tests still failing due to a known limitation with caller(0) returning empty warning bits in main script context (documented in lexical-warnings.md)
• sub-defer.t and sub-quote.t: Timeout/OutOfMemoryError - this is a pre-existing issue that occurs even without any changes to WeakRefRegistry or ScalarUtil

OOM Investigation

The OOM error in sub-defer.t and sub-quote.t was investigated extensively:

• Confirmed to be pre-existing (occurs even without any changes)
• Tried optimizing clearAllBlessedWeakRefs by filtering during snapshot to reduce memory pressure - did not help
• Tried adding PJ_SKIP_WEAK_CLEAR environment variable to skip weak ref cleanup - did not help
• Tried increasing heap size with JPERL_OPTS="-Xmx4g" - still failed with OOM
• The tests pass all subtests (33 for sub-defer.t, 51 for sub-quote.t) but then timeout during cleanup
• The OOM is thrown from the UncaughtExceptionHandler in the orphan-watchdog thread

Root cause identified:
The OOM occurs in ScalarRefRegistry.forceGcAndSnapshot() at line 130, which is called by ReachabilityWalker.sweepWeakRefs() at line 1120 during cleanup. The forceGcAndSnapshot method runs 3 passes of GC (each calling System.gc() 5 times with 10ms sleep) to determine which objects are still reachable. With the many CODE refs and weak refs created by Sub::Defer/Sub::Quote, this GC forcing mechanism runs out of memory even with 4GB heap.

The Sub::Defer/Sub::Quote tests create many CODE refs and weak refs, and during cleanup, the reachability walker's GC forcing mechanism causes OOM. This is a fundamental part of the selective refcounting system and not easily fixable without redesigning that system.

Potential fixes (not implemented in this PR):

• Skip or reduce GC cycles in forceGcAndSnapshot for memory-intensive scenarios
• Add environment variable to disable reachability sweeping for specific tests
• Accept as known limitation - these tests stress the GC beyond normal limits

Files Changed

• WeakRefRegistry.java: Changed clearWeakRefsTo to always include CODE refs