Add an ASCII fast path for String#casecmp?

[sampokuokkanen]

String#casecmp? folds both operands with a Unicode case fold, allocating two new strings, and compares the results. For ASCII-only strings, Unicode case folding is equivalent to ASCII case-insensitive comparison, so the operands can be compared byte-by-byte without allocating anything.

#casecmp is older and does not take Unicode into account, which makes it faster. So this will add a fast path to #casecmp? so that the performance will be around the same. Turned off GC for that benchmark as it made the bench noisy (like in ruby/benchmark/string_concat.yml).

Benchmark

Apple M1 Pro (arm64-darwin)

case	master	this PR	speedup
casecmp_p-1 (36 B)	3.12M	18.57M	~5.9x
casecmp_p-10 (360 B)	358k	3.00M	~8.4x
casecmp_p-100 (3.6 KB)	282k	357k	~1.3x
casecmp_p-1000 (36 KB)	25.4k	39.3k	~1.5x

Docker Linux (aarch64-linux)

case	master	this PR	speedup
casecmp_p-1 (36 B)	5.03M	18.47M	~3.7x
casecmp_p-10 (360 B)	744k	2.73M	~3.7x
casecmp_p-100 (3.6 KB)	130k	379k	~2.9x
casecmp_p-1000 (36 KB)	12.1k	39.3k	~3.2x

AMD Ryzen 7 5700G (8C/16T) — Windows 11 (x64-mingw-ucrt, gcc 16.1.0, MSYS2 UCRT64)

case	master	this PR	speedup
casecmp_p-1 (36 B)	6.38M	18.32M	~2.9x
casecmp_p-10 (360 B)	1.20M	3.18M	~2.7x
casecmp_p-100 (3.6 KB)	175k	348k	~2.0x
casecmp_p-1000 (36 KB)	15.97k	35.11k	~2.2x

Non-ASCII (fast path not taken) is unchanged:

case	master	this PR	speedup
casecmp_p-nonascii1	1.221M	1.235M	~1.0x
casecmp_p-nonascii10	161.96k	159.39k	~1.0x
casecmp_p-nonascii100	16.86k	16.81k	~1.0x
casecmp_p-nonascii1000	1.749k	1.753k	~1.0x