[C++][Parquet] Implement Float16 logical type

[benibus]

Describe the enhancement requested

There's currently an active proposal to add a half-float logical type to the Parquet spec here: apache/parquet-format#184. Following the discussion in the PR/issue, the general consensus was that we should go ahead and implement support for the type before moving forward with a vote.

To summarize, this would currently entail adding a float16 logical type (based on a 2-byte fixed-size binary physical type) and implementing read/write support. We also want to ensure that its ordering requirements are consistent with the native floating point types, min/max values are properly handled in Statistics, etc.

Component(s)

C++, Parquet