[C++][Parquet] Optimize delta bit-packed decoding when bit-width = 0

[pitrou]

Describe the enhancement requested

DELTA_BINARY_PACKED decoding has limited performance due to a back-to-back dependency between the computations of value N and value N+1.

However, that dependency theoretically disappears if we know that all deltas are 0, i.e. when the delta bit width is 0. In this case (which can occur in some miniblocks, depending on input structure), decoding can become much faster as loop iterations can progress in parallel.

Component(s)

C++, Parquet