Due to a terminating NUL that was placed in a `r#` string, we were not
actually NUL-terminating pipe names on Windows. While this has no
security implications due to the random nature of the prefix, it would
occasionally cause random failures when the trailing garbage would make
the pipe name invalid.
Investigating https://github.com/denoland/deno/issues/22574
Unable to reproduce with a unit test, but assuming that it's a name
collision or create pipe/open pipe race, and adding some additional
diagnostics.
