keyword padding optional resp. hash-specific
for the 3 new MPH's (chm, chm3, bpz) we use a modified jenkins hash from NetBSD, mi_vector_hash. this is not a particularly good hash function, and requires padding to 4 byte boundaries. we could switch to wyhash (as done in my nbperf), or the shorter fnv1a (esp. good for short keys).
and only with the current jenkins add the padding bytes. the problem is that size-optimized hashes, such as bpz could be used on size-constrained environments, eg. with -N NONE (without lookup table and no manual false-positive checks), but this would then add the potential -Os padding overhead of the keywords. avr does not align the wordlist members.