arrow

ARROW-9197: [C++] Overhaul integer/floating point casting: vectorize truncation checks, reduce binary size

Bunch of stuff in this PR:

* Speed up safe integer/floating<->integer/floating casts, especially when they are mostly not null
* Compiled size of scalar_cast_numeric.cc.o is down to 736KB from 1210KB. There's about 70KB of new code in util/int_util.cc.o for some of the stuff below. Net reduction of 400KB in libarrow.so on Linux/clang-8
* Start overdue casting benchmark suite. There are benchmarks for some of the casts I worked on in this PR to help show the before/after performance
* General purpose `CheckIntegersInRange` for fast range-checking of integer arrays
* Augment scalar-cast-test to check casting `arrow::Scalar` values, too. I disabled these casts for the types where they aren't supported: decimal (see ARROW-9194), string (ARROW-9198), dictionary, lists (ARROW-9199), and temporal types (ARROW-9196).
* I discovered by `nm` sleuthing that our code in `BoxScalar` in codegen_internal.h was generating a lot of binary for some reason, so this has been fixed.

Closes #7506 from wesm/ARROW-9197

Authored-by: Wes McKinney <wesm@apache.org>
Signed-off-by: Wes McKinney <wesm@apache.org>