remove the use of String's len field (introduced in #22133)

[rfourquet]

The len field has been removed in 2b47865.
I assume here that jl_alloc_string (called by _string_n) reserves one char more for \0 that some library will insert, as far as I can understand its code (it seems len+1 bytes are allocated, where len is the requested size, the last byte being zeroed).

[yuyichao]

Strings shouldn't be mutated like this. Use StringVector instead.

[timholy]

I went ahead and canceled a bunch of CI runs on AppVeyor and Travis, since nothing will pass until this gets through.

[JeffBezanson]

This must mean there are no tests for this function? Could you add some?

[JeffBezanson]

I see, I guess there are tests. I didn't realize this code changed so recently.

[rfourquet]

Strings shouldn't be mutated like this.

I thougth this was the purpose of _string_n (it's used like this in "serialize.jl" and "strings/string.jl", as far as I can tell...)

[rfourquet]

replaced _string_n by StringVector

Review comment was addressed.

[timholy]

@rfourquet, thanks for taking the lead on this. These kinds of "distributed errors" aren't anyone's fault and are unavoidable when changes are happening at a rate that exceeds individual-human bandwidth. Having someone who takes ownership and deals with it so quickly is simply priceless.