Tune translate_cds implementation

[holtgrewe]

A lot of time in mehari is consumed in the translate_cds implementation.

We should properly tune and benchmark this function in hgvs-rs.

[holtgrewe]

cargo bench result for current phf implementation

[holtgrewe]

After switching to byte arrays with builtin match (two changes, not 100% clear where time is added/lost).

[holtgrewe]

Note: converting to a hard-coded trie (nested match) does not help. Creating result with capacity helps a bit (5%).

[holtgrewe]

Even faster, with fast hash maps

[holtgrewe]

Changing everything to bytes makes the interfaces painful.

I found an implementation that will allow to keep interfaces using str and String but still gives a speedup of almost 4x.