[C++][Parquet] Generalize bpacking and possible speedup

[AntoinePrv]

Describe the enhancement requested

Following GH-47514 we can safely try to improve the bpacking functions:

• Adding support to unpack to a 16bit integer type (and investigate speedup in BitPacked decoder).
• Uniform const uint8_t* in API.
• Reduce the amount of code generated through some factorization.
• Investigate low hanging fruit speedup to unpack functions (xsimd::batch_constant...).

Component(s)

C++, Parquet

[pitrou]

Issue resolved by pull request 47573
#47573