docs: complete feature compatibility audit

[fglock]

Summary

• Complete the documented feature-gap audit started in PRs docs: add PerlOnJava feature audit plan #993 and docs: continue feature compatibility audit #994.
• Add isolated probes for unicode_strings, multibyte encoded I/O, bigint,
  bignum, and bigrat.
• Reconcile every remaining feature-matrix entry into probe evidence,
  representative partial support, or an explicit source/JVM boundary.
• Synchronize the feature matrix, roadmap, audit plan, and rerunnable probe README.

Key findings

• unicode_strings, bigrat, and representative multibyte seek/tell/
  truncate behavior pass on native Perl and both PerlOnJava backends.
• bigint loses precision in the interpreter; the interpreter bignum probe
  times out. The JVM backend passes both basic probes.
• Remaining unsupported features such as restricted hashes, fork, DBM, and
  ops have explicit native-vs-backend evidence.
• Debugger, JSR-223, XS, and true-fork gaps are classified using their current
  implementation/design sources or JVM boundary constraints.

Verification

• Native Perl validation was run before every new probe was used as an oracle.
• New probes were run with both ./jperl and ./jperl --interpreter, each with
  a timeout and captured output.
• make check-links passed: 575 links checked, 0 errors.
• The full build/test suite was not run; this work changes audit probes and
  documentation only.

Generated with Codex (https://openai.com/codex/)

Co-Authored-By: Codex noreply@openai.com