arrow

ARROW-4752: [Rust] Add explicit SIMD vectorization for the divide kernel

Adds SIMD support for divide binary op.  There are probably a few more optimizations to be made but the performance of divide is now very close to the other binary ops on my computer which is pretty good considering all the extra checking involved to allow for divide by zero correctly.

PTAL

Closes #5143 from paddyhoran/compute-divide and squashes the following commits:

71a5a31aa <Paddy Horan> Renames utility function and makes it more flexible.
f63a41039 <Paddy Horan> Adds testing for `simd_load_without_invalid_zeros`.
98189212d <Paddy Horan> Adds test for `is_valid` utility function.
0d221edad <Paddy Horan> Cleans up docs and trait bounds.
66580c21a <Paddy Horan> Minor fixes.
baee6d5de <Paddy Horan> WIP for SIMD support for divide op.

Authored-by: Paddy Horan <paddyhoran@hotmail.com>
Signed-off-by: Andy Grove <andygrove73@gmail.com>