[Lex] Keep track of skipped preprocessor blocks and advance the lexer directly...

[Lex] Keep track of skipped preprocessor blocks and advance the lexer directly if they are revisited

This speeds up preprocessing, specifically for preprocessing the clang sources time is reduced by about -36%,
using measurements on M1Pro with a release+thinLTO build.

Differential Revision: https://reviews.llvm.org/D127379