[PATCH] Make memset in calloc a tail call.

[Roland McGrath]

That seems fine.  There is no reason not to do it even if the compiler
(eventually) would have figured it out.  

I also don't see any reason to keep the hand-unrolled code for memset < 8.
The compiler and/or memset implementation should optimize that case well
enough.  Of course that is an orthogonal change, and it should be verified
that the generated code does not get manifestly worse performance.