Refactor scalar reductions to use common execution policy

[jjwilke]

Following the design doc for improving code reuse, this implements all scalar reduction kernels through a single implementation.

• Study verifies binary size is not increased
• Study verifies performance is equivalent

[magnatelee]

@jjwilke please add a copyright header to the new files

[magnatelee]

left two nitpick-y comments. otherwise looks good.

[jjwilke]

@magnatelee Should be ready to merge. There was an additional change required to unary_red_util.h you might want to glance at.

[manopapad]

Missing newline at end of file. Weird that the clang auto-formatter didn't catch that. Does it look like clang-format is running when you make change to header files? It might be that the filename pattern is skipping header files, or maybe just our clang-format configuration isn't checking for missing newlines at end of files.

[magnatelee]

@magnatelee Should be ready to merge. There was an additional change required to unary_red_util.h you might want to glance at.

Feel free to merge this once you remove the obsolete header file.