Performance regression in skipchars

[KristofferC]

skipchars took a 5x preformance hit (for the benchmark we have in BaseBenchmarks) presumably after the introduction of locks on the write calls.

Perhaps we can lock outside of the loop and write without a lock and finally unlock it when we are done?

[StefanKarpinski]

cc @vtjnash

[vtjnash]

Duplicate of #34195 / addressed in #35426. We might be able to do some more automatic compiler optimization of this in the future, instead of the optimization provided there, though it breaks some forward-progress guarantees required by that code pattern in general, so remains to be investigated.