Speed up pad_nulls for FixedLenByteArrayBuffer (#6297)

* optimize pad_nulls for fixed_len_byte_array

* start refactor of benchmarks

* refactor float16

* add fixed_len_byte_array to float16 benches

* replace copy_within with vectorizable copy

* update comment

* move branch on byte_length outside of loop

* reduce code duplication while preserving performance gains

* formatting

* silence clippy

* clippy won again