Minor optimization opportunity (ARM NEON, minify)

Currently the minifier under ARM NEON does a little bit too much work by doing part of the work necessary to identify structural characters as well as all of the work necessary to identify white spaces. This is in contrast with x64 where these are identified using distinct code paths.

Our minifier even on ARM NEON is likely the fastest out there, so this is a minor concern that we choose to ignore deliberately.

It would be easy to at least quantity the effect by hacking the code.

Note: This only affects ARM processors.