Improve Unicode regex full-fold parity

[fglock]

Summary

• derive the complete multi-character case-fold table from the ICU4J runtime dependency
• recognize full folds in escaped literals, reverse-fold sequences, and simple character classes
• preserve Perl's distinct /a and /aa compiler flags and avoid rewriting across scoped (?aa:...) groups
• add system-Perl-validated regression coverage for sharp-s, long-s, ligatures, classes, and /a versus /aa

Fold results

Test	Before	After	Change
re/reg_fold.t	6663/7111	6925/7111	+262 passing
re/fold_grind_u.t	66/68842	66/24309	44,533 fewer emitted failures
re/fold_grind_d.t	22/11469	22/7591	3,878 fewer emitted failures
re/fold_grind_a.t	21/8609	26/2147	+5 passing, 6,462 fewer emitted failures
re/fold_grind_aa.t	18/25283	75/22811	+57 passing, 2,472 fewer emitted failures

The grind files summarize successful internal subtests but emit individual failures, so a lower denominator means additional internal cases now pass silently. op/ref.t remains unchanged at 421/481 and is outside this fold-focused change.

Validation

• make — passed, all unit shards (5m12s)
• prove src/test/resources/unit/regex_full_casefold.t with system Perl — 8/8 passed
• six-file perl_test_runner.pl comparison — no timeouts; fold counts above