Optimise StrExt::replace_smolstr, StrExt::replacen_smolstr for single ascii replace

[alexheretic]

This PR ports a single ascii fast path optimisation from std str::replace highlighted by benchmarks in #96. I also made it a little more general applying to both StrExt::replace_smolstr & StrExt::replacen_smolstr.

PR bench

group                       faster-1-ascii-replace    master
-----                       ----------------------    ------
replace_smolstr len=12      1.00      6.8±0.03ns      3.51     23.7±0.27ns
replace_smolstr len=50      1.00     29.3±0.05ns      2.84     83.1±0.04ns
replace_smolstr len=1000    1.00    200.2±0.01ns      3.17    635.1±5.58ns

The change provides a significant perf improvement particularly for inline reprs. Also faster than String for len=12 now 🙂 (I measured previously as 10.8ns).

[Veykril]

Thanks!