Add portable replacement tests for excluded XS::APItest bigmem stack tests

The issue requires adding new tests to src/test/resources/unit. First, examine the excluded upstream tests perl5_t/t/bigmem/stack.t and perl5_t/t/bigmem/stack_over.t to understand the portable behaviors they represent. Then, design new Perl-level tests that validate list-context propagation, argument counts, list operations, and bounded stack behavior without XS::APItest. Ensure tests pass under system Perl, the JVM backend, and the interpreter backend. The acceptance criteria include running within CI limits and documenting the exclusions.

Summary

Add PerlOnJava-owned replacement coverage for the portable behavior represented
by the Perl core tests:

• perl5_t/t/bigmem/stack.t
• perl5_t/t/bigmem/stack_over.t

Those upstream tests are now excluded from dev/tools/perl_test_runner.pl
because they require the native XS::APItest extension and directly test
Perl 5 C-level value-stack/MARK implementation details. They also require
multi-gigabyte allocations that are unsuitable for the PerlOnJava test suite.

The exclusion is intentional, but it must not remove all coverage of the
observable list, argument-stack, and bounded overflow behavior that is
portable across Perl implementations.

Current upstream-test classification

bigmem/stack.t depends on XS::APItest functions including wide_marks,
xs_items, eval_sv, and call_sv, and creates arrays with indices beyond
0x8000_0000. It exercises native mark-stack offsets, XS argument counts,
large list operations, split, repeat, tie, and related C implementation paths.

bigmem/stack_over.t also depends on XS::APItest and specifically checks
32-bit MARK behavior on a 64-bit native Perl build while allocating an array
with a 2-billion-element index.

These implementation-specific assertions cannot be ported verbatim to the
JVM or interpreter backend.

Proposed replacement coverage

Add focused tests under src/test/resources/unit for Perl-level behavior that
does not depend on XS, native pointers, or enormous allocations. At minimum,
cover:

1. List-context propagation through nested calls and expressions.
2. Correct item counts for bounded list construction, list assignment, and
   list-returning functions.
3. List slicing and negative indexes after a list-producing call.
4. split, join, grep, and repetition with bounded inputs.
5. Argument counts observed by ordinary Perl subs through @_, including
   nested calls and calls returning multiple values.
6. eval and require paths that occur while a list/argument stack is active.
7. Bounded recursive or deeply nested calls that verify the runtime reports a
   controlled stack/resource failure rather than hanging or corrupting values.

Use sizes large enough to expose indexing and context bugs but small enough for
ordinary CI machines. Do not reproduce the 2-billion-element allocation or the
60/17 GB memory requirements.

Test requirements

• Validate every new Perl-level test with system Perl first.
• Require success on both JVM and interpreter backends.
• Keep the tests deterministic and independent of XS::APItest.
• Do not weaken or modify the upstream Perl core tests.
• Preserve the runner exclusions for the two native-only upstream files.
• Add comments documenting which native XS::APItest assertions are
  intentionally not represented by the portable replacements.

Acceptance criteria

• A project-owned test suite covers the portable semantics listed above.
• The suite passes under system Perl, PerlOnJava JVM, and PerlOnJava
  interpreter execution.
• The tests run within normal CI memory and time limits.
• The compatibility documentation or test-runner notes explain why the two
  upstream XS tests remain excluded.