Experimental "slab" allocator for ops.

Experimental "slab" allocator for ops.
To try it -DPL_OP_SLAB_ALLOC for op.c
This is for proof of concept only, it leaks memory
(ops are not free'd) so don't use in embedded apps.
If this minimalist version does not show performance
gain then whole idea is worthless.
Nick see's approx 12% speed up vs perlmalloc running
  perl -Ilib -MCPAN -e ''
Solaris2.6, gcc-2.8.1 but numbers are not repeatable.

p4raw-id: //depot/perl@3424