Optimise in(::Char, ::Union{String, SubString{String}}) (#62607)
Optimise this slightly by inlining, which eliminates some dead code. For
short strings, this is ~60% faster.
Supersedes https://github.com/JuliaLang/julia/pull/43989.
Closes #43980.
Benchmark:
```
s = "chalet"
@btime in('\0', $(Ref(s))[])
@btime in('t', $(Ref(s))[])
s = "h"^1000*'t'
@btime in('\0', $(Ref(s))[])
@btime in('t', $(Ref(s))[])
```
Timings, before / after:
```
7.410 ns / 4.097 ns
7.713 ns / 4.097 ns
14.858 ns / 8.695 ns
14.808 ns / 8.755 ns
```
There are several other strategies discussed in that PR and that issue,
but I believe this is a generally better strategy.
Using `memmem` is infeasible because checking for valid utf8 in the char
takes too long. It may be better for long haystacks, though. In testing,
the haystacks have to be _quite_ long before it's worth it- like,
thousands of bytes. You could add a bunch of fast paths, but most of
these are already implemented in `findfirst(::Char, ::String)`.
