bench: zero buffers before copying them

[gperciva]

Nitpick for static analyzers. Currently we're calling kvldskey_create() on uninitialized buffers, and then initializing the buffer afterwards. Copying uninitialized bytes around smells fishy.

This does a memset() before kvlds_create().

I don't see any impact on performance using a smaller version of the local benchmark suite:

Looks like statistical noise to me. I ran the benchmark 11 times (discarding the first run); here's some sample output from one time:

bulk_insert: 4096 408619
bulk_insert: 8192 568494
bulk_insert: 16384 568948
bulk_insert: 32768 594021
bulk_insert: 65536 605229
bulk_update 4096 392512
bulk_extract 4096 759357
random_read 4096 954925
random_mixed 4096 270577
hotspot_read 65536 948214