Reuse Unicode case analysis when normalizing name parts

[iliaal]

Summary

• Analyze the case shape once when detecting mixed-case name parts, instead of calling upper/lower predicates that each repeat the same Unicode analysis.
• Preserve the existing 1,024-byte guard, combining-mark stripping, case-transition rule, and long-token fallback.
• Add six regression cases covering uppercase, uncased scripts, an uncased prefix, decomposed accents, punctuation, and empty input.

Benefit

Removes a redundant letters extraction and upper/lower conversion for non-uppercase parts within the guard. A focused PHP 8.4.26 benchmark normalizing 400,000 fresh surname parts across eight ASCII/Unicode inputs took 804–909 ms before and 684–757 ms after in five paired runs. This is a normalization microbenchmark, not an end-to-end parser throughput claim.

Validation

• PHPUnit: 1,203 tests / 3,109 assertions passed with the 256 MiB limit.
• PHPStan level 9 and PHP 8.5 compatibility analysis passed.
• PHP CS Fixer dry run, 69 PHP syntax checks, Composer audit, baseline/README/deprecation guards, and diff checks passed.
• Independent review found no behavior regressions.
• All 84 remote file entries (paths, modes, blob hashes) match the tested tree, verified by complete NUL-delimited Git tree comparison before PR creation. Tree: 1983afc3fe3659f81f9ffaa8efed47de20538491.

Exact-head CI passed all five jobs: PHP 8.3 / 8.4 / 8.5 tests (including Composer audit), static analysis (including the PHP 8.5 compatibility gate), and code style. No jobs or steps were skipped.