segfault in test-strcmp

[Roland McGrath]

> The test should probably just test defined behaviour, so
> should terminate all strings, or at least have an option whether it allows
> accesses past the place where string differs or where searched char was
> found (by default only testing with zero terminated strings).
> Will look into this.

Yes, I don't think we want to test the undefined behavior.  We don't want
to make it part of GNU behavior that you can call strcmp on unterminated
strings even if they differ in the first character.