arrow

PARQUET-1840: [C++] Stop Early on DecodeSpaced

This only makes appropriate stoppage early when possible:
1.  When there are no more values to use.
2.  When all remaining elements are present.

I put a TODO to look at SIMD for using intrinsics to potentially speed up moving the data (when they can be applied for ints/floats).  I tried to do this without SIMD and it looks like it made things slightly worse, so I removed it.  @jianxind this might be of interest to you.

As it stands these changes seem to be within noise on the current benchmarks but I expect in real world scenarios they could improve a a decent amount (mostly populated or mostly null values).

Closes #6908 from emkornfield/double_write

Lead-authored-by: Micah Kornfield <emkornfield@gmail.com>
Co-authored-by: Antoine Pitrou <antoine@python.org>
Signed-off-by: Antoine Pitrou <antoine@python.org>