On large values, this consumes ~half the memory.

This looks like a correct, comprehensible change, my only wonder would be whether some sort of regex-based approach would be more efficient for this quote direction as well, i.e., have a regex recognize strings of safe characters.

It may also depend on what sorts of strings are typically quoted/unquoted. Different texts/corpuses probably have different performance characteristics, and I am not all that familiar with how this gets used in the wild.

there are likely further optimization opportunities. but as this does use notably less memory already, lets go ahead and do it. I don't like introducing regexes when possible that could be tested and considered as a separate change if anyone ever cares here.