Use SearchValues for identitifers

[sebastienros]

No description provided.

[sebastienros]

@lahma wanted to compare raw "Contains" usage. Then expanded with the current code path that uses it to parse an identifier in a loop checking each char, vs the vectorized version that won't use a loop.

From that I will try to remove the usage of the char map (for net80+) and run the higher level benchmarks

[lahma]

In need to gather my thought with this probably some morning coffee, but conceptually looks good. It's fast with modern runtimes but legacy not so much, I don't mind though and Parlot is plenty fast anyway.

[sebastienros]

Added benchmarks specifically for SkipWhiteSpace slightly changedthe current implementation. Vectorization would help when there are bigger spans of spaces. A mathematical expression might not matter, but it could have a huge impact in things that use spaces as indentation (JSON, source code ...).

[sebastienros]

I had this idea that languages where it could matter would include new lines in their separators (liquid) so I benchmarked it with the vectorization or not and couldn't see a difference either.