bug#9612: sort: avoid a NaN-induced infloop

[Jim Meyering]

Jim Meyering wrote:
> Who would have thought that including a few NaNs in the input to sort
> would make it infloop.
>
> The original failure arose only when sort was reading from a pipe:
>
>     yes -- -nan | head -156903 | sort -g > /dev/null
>
> But it's not always easy to reproduce.
...
> [ I'm about to file the glibc bug
...

I've done that:

  RFE: strtold: do not include uninitialized bytes when converting "NaN"
  http://sourceware.org/bugzilla/show_bug.cgi?id=13246