Two micro-opts for lexer

[tigercat2000]

Measured about ~150ms faster peek() with the first change.
Measured within-noise for the second change- but it's nicer code so it's still a good change, I think.

[SpaceManiac]

150ms must be a typo; did you mean ns? The speedup is within noise parsing an entire codebase on a release build. It seems to me that on -C opt-level=2 or higher, where PartialEq gets inlined, the assembly has a len== check anyways.

Improving the put_back code is good, but you could go even further by moving the let ws = ... inside read_ident and return bool instead of Option<u8>.