[C++] Benchmark key_hash and document tradeoffs with vendored xxhash

[asfimport]

ARROW-15239 adds a vectorized hashing function for use initially in execution engine bloom filters and later in the hash-join node. We should add some benchmarks to explore how the performance compares to the vendored scalar xxhash implementation. In addition, we should document where the two differ and explain any tradeoffs for future users.

Reporter: Weston Pace / @westonpace

Related issues:

• [C++][Compute] Add scalar_hash function (relates to)

_{Note: This issue was originally created as ARROW-16017. Please see the migration documentation for further details.}

[asfimport]

Aldrin Montana / @drin:
Linking to ARROW-8991, where the PR starts to add some of this benchmarking