rdrand_rng() asm loop body executes unconditionally: called with limit < 8 it writes a full 8-byte RDRAND word past a sub-word buffer

Summary

do-while instead of while: any limit in [0,7] yields word count 0 yet the body stores 8 bytes. Sole in-tree caller passes 16 into a 48B buffer (in-bounds) - latent hardening for the next caller. test-rcx-zero guard fix in the .S.