fix: bad usage of std::string_view::npos

[lemire]

Alternative to #331

credit to @codebytere

[lemire]

@anonrig It is not clear to me that we can come up with a test for this because it is not really a buffer overflow (or, at least, the possibility of a buffer overflow has not been demonstrated). Rather the issue is that the code would create an enormous std::string_view pointing at an insane range... but this range would not be accessed so we were fine. It is still terrible code that needs fixing... and it can, indeed, trigger an assert as in llvm/llvm-project#61100

[lemire]

It is pretty obviously bad code that needs fixing. It is not a nuanced thing... I think it just escaped our attention so far because it is a relatively niche scenario.

[anonrig]

thanks @codebytere

[lemire]

@anonrig commit amended to stipulate that @codebytere is co-author.