Increase PerlOnJava test pass rate

[fglock]

Increase PerlOnJava test pass rate

Use the 2026-08-24 Perl test-runner report as the baseline:

• 697,285 tests total
• 686,978 passing (98.5%)
• 10,317 not passing
• 1,012 tests blocked by incomplete runs
• 378/623 files fully passing

The priority is meaningful Perl compatibility and crash-unblocking work. Environment-only, XS-dependent, Perl-internals, and optree tests should be tracked separately rather than treated as runtime compatibility regressions.

Phase 1: unblock incomplete runs

Fix the highest-value fatal failures first:

• op/write.t: format lookup currently fails with Undefined format "main::OUT".
• op/const-optree.t: Not a CODE reference handling.
• op/goto.t: interpreter null-result handling during nested dynamic goto.
• op/kvhslice.t: hash-slice mutation/read-only behavior.
• op/rt119311.t: destruction and call-stack unwinding through END blocks and formats.
• mro/package_aliases*.t: package-stash alias invalidation and MRO lookup.
• comp/form_scope.t: closure/form handling.
• op/decl-refs.t: declared-reference parsing and runtime behavior.
• comp/parser.t: parser error recovery and diagnostic compatibility.

Target: eliminate the fatal termination in these files and recover most of the 1,012 blocked tests.

Phase 2: fix concentrated semantic clusters

• Tie dispatch: correct FETCH/STORE counts and tied-glob behavior in op/tie_fetch_count.t, op/tie.t, and op/tiehandle.t.
• Signatures/prototypes: prioritize binding, defaults, slurpy parameters, and diagnostics in op/signatures.t and related tests.
• File tests and I/O: address op/filetest.t, op/utf8decode.t, and remaining io/* failures.
• Comparison chains: fix the shared compiler/runtime issue behind op/cmpchain.t.
• Package/version parsing: fix the shared parser behavior behind op/packagev.t.
• Smartmatch: improve supported scalar, reference, array, hash, and coderef cases in op/smartmatch.t.

Phase 3: diagnostics and regression coverage

After behavior is correct, normalize Perl-compatible error categories, wording, source locations, and UTF-8 rendering, especially for lib/croak.t and parser diagnostics.

For every externally observed failure:

1. Add a focused project-owned regression test.
2. Validate expected behavior with system Perl first.
3. Verify JVM and interpreter backends where applicable.
4. Preserve existing Perl core tests unchanged.

Acceptance criteria

• Run affected tests with dev/tools/perl_test_runner.pl, hard timeouts, and captured output.
• Run focused unit tests and the full make gate after each cluster.
• Track passing tests, incomplete tests, fully passing files, and regressions separately.
• First milestone: recover approximately 800–1,000 blocked tests without introducing new failures.

Baseline source: logs/test_20260824_200000_1101.log, generated by dev/tools/perl_test_runner.pl.