perf(tspath): void []byte allocation in ToFileNameLowerCase

[camc314]

Replace for _, c := range []byte(fileName) with direct byte indexing to iterate over the string without allocating a new []byte slice. This removes an unnecessary allocation in both the ASCII fast path check and the lowercasing loop, improving performance and reducing GC pressure.

                                                     │ bench-to-file-name-lower-case-main.txt │ bench-to-file-name-lower-case-cbdcb60a5.txt │
                                                     │                 sec/op                 │        sec/op          vs base              │
ToFileNameLowerCase//path/to/file.ext-12                                         8.422n ± ∞ ¹            8.508n ± ∞ ¹       ~ (p=0.151 n=5)
ToFileNameLowerCase//PATH/TO/FILE.EXT-12                                         38.27n ± ∞ ¹            38.27n ± ∞ ¹       ~ (p=0.730 n=5)
ToFileNameLowerCase//path/to/FILE.EXT-12                                         41.08n ± ∞ ¹            41.03n ± ∞ ¹       ~ (p=0.421 n=5)
ToFileNameLowerCase//user/UserName/proje...etc-12                                58.59n ± ∞ ¹            58.32n ± ∞ ¹       ~ (p=0.841 n=5)
ToFileNameLowerCase//user/UserName/proje...etc#01-12                             161.1n ± ∞ ¹            161.5n ± ∞ ¹       ~ (p=0.635 n=5)
ToFileNameLowerCase//user/UserName/proje...etc#02-12                             155.1n ± ∞ ¹            154.1n ± ∞ ¹       ~ (p=0.595 n=5)
ToFileNameLowerCase//user/UserName/proje...etc#03-12                             139.5n ± ∞ ¹            136.7n ± ∞ ¹  -2.01% (p=0.048 n=5)
ToFileNameLowerCase/FoO/FoO/FoO/FoO/FoO/...etc-12                                550.8n ± ∞ ¹            548.7n ± ∞ ¹       ~ (p=0.841 n=5)
geomean                                                                          78.82n                  78.58n        -0.30%

[camc314]

@jakebailey up to you if this is merged, I thought this had a perf benifit, but i explored the compiled code in godbolt, and it appears exactly the same machine code is generated .

Regardless, i think this is slightly more explicit behaviour.

[jakebailey]

Yeah, this kind of conversion I am not surprised does not affect the compile. I agree it's clearer.