This is a follow up task of
ORC-1020 which didn't optimize the code path when the run has nulls. Adding a buffer to decode the whole run at once can leverage the improvement of ORC-1020. Not just for the DIRECT encoding, this also benifits other encodings like PATCHED_BASE and DELTA. It also helps to remove the state variables in RleDecoderV2 and improves the code redability.
- links to