Optimize some inner cases for tokenFromKeyword

[MaxGraey]

Some nested cases which contain a lot of comparisons (>= 5) with long keyword lengths and common prefixes could be optimized more. WDYT?

benchmark
Parse AS.
main branch:

Parse      :   785.185 ms  n=171
Parse      :   768.248 ms  n=171
Parse      :   801.483 ms  n=171
Parse      :   776.191 ms  n=171

This PR:

Parse      :   727.563 ms  n=171
Parse      :   759.355 ms  n=171
Parse      :   760.821 ms  n=171
Parse      :   769.710 ms  n=171

• I've read the contributing guidelines
• I've added my name and email to the NOTICE file