Optimise to_lowercase_smolstr, to_uppercase_smolstr

[alexheretic]

Optimisation opportunity highlighted by testing against String perf and using benchmarks in #96.

This is slightly more involved than the ascii optimisations in #97. I ported std convert_while_ascii into an inline version inline_convert_while_ascii to get an equivalent ascii-optimistic optimisation to the String version.

PR bench

group                                  master                  faster-case
-----                                  ------                  -----------
to_lowercase_smolstr len=12            2.13     65.3±0.20ns    1.00     30.7±0.03ns
to_lowercase_smolstr len=50            5.70    354.6±0.43ns    1.00     62.2±0.03ns
to_lowercase_smolstr len=1000          52.06     5.8±0.01µs    1.00    111.6±0.06ns

The change provides a significant perf improvement on inline & heap reprs. Also faster than String for len=12 now 🙂.

The benchmark is favourable to this optimisation as it measures a string with mostly ascii prefix. But I think this is a very common case for this method so the optimisation makes sense generally, and clearly std think so too.

[alexheretic]

@Veykril any issues with this? It would be nice to get these, and the other pr, improvements into a new smol_str release.

[Veykril]

Ah sorry, forgot about this, I'll try to get back to this + do a release the coming week as I am a bit busy right now (if I'm not coming back to this this week please aggressively re-ping me)