perf: improve case-insensitive string compare

[francisdb]

Did some performance testing and this was a serious bottleneck for files with large amounts of streams.

master:

write many smaller streams/total/default
                        time:   [75.344 ms 75.390 ms 75.472 ms]
                        thrpt:  [828.12 KiB/s 829.02 KiB/s 829.53 KiB/s]

this pull request:

write many smaller streams/total/default
                        time:   [9.8702 ms 9.9169 ms 9.9802 ms]
                        thrpt:  [6.1156 MiB/s 6.1547 MiB/s 6.1838 MiB/s]
                 change:
                        time:   [−86.927% −86.889% −86.837%] (p = 0.00 < 0.05)
                        thrpt:  [+659.70% +662.75% +664.96%]
                        Performance has improved.

I have even more improvements coming but wanted this merged first.

[mdsteele]

Thanks for digging into this!

[francisdb]

Thanks for the review. I can add a benchmark specifically for these calls if you want. But maybe that's a bit too targeted. The main benchmark already shows a huge regression on the ASCII part.

Anyway. I will double check the improvement on the non-ascii case as I am also a bit unsure if that change is worth it. -> NO, reverted.

[mdsteele]

Looks good, thanks