Introduce PrimitiveValueDecoder to enable batch decoding of values

[Jefffrey]

Though the benchmarks are kinda simple, they identified (via flamegraph) that there was a significant amount of time being spent on next() calls in RleReaderV2. So introduce a new PrimitiveValueDecoder to enable batch decoding support, currently only properly implemented for integer arrays with no null buffer. Even so, benchmarks show significant improvement:

datafusion-orc$ cargo bench
   Compiling orc-rust v0.3.1 (/home/jeffrey/Code/datafusion-orc)
    Finished `bench` profile [optimized + debuginfo] target(s) in 11.65s
     Running unittests src/lib.rs (/media/jeffrey/1tb_860evo_ssd/.cargo_target_cache/release/deps/orc_rust-53e462447a03afd6)

...

     Running benches/arrow_reader.rs (/media/jeffrey/1tb_860evo_ssd/.cargo_target_cache/release/deps/arrow_reader-419fc98a1d304edb)
Benchmarking sync reader: Warming up for 3.0000 s
Warning: Unable to complete 100 samples in 5.0s. You may wish to increase target time to 8.0s, or reduce sample count to 60.
sync reader             time:   [79.523 ms 79.618 ms 79.721 ms]
                        change: [-42.234% -42.072% -41.919%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 6 outliers among 100 measurements (6.00%)
  5 (5.00%) high mild
  1 (1.00%) high severe

Benchmarking async reader: Warming up for 3.0000 s
Warning: Unable to complete 100 samples in 5.0s. You may wish to increase target time to 8.0s, or reduce sample count to 60.
async reader            time:   [80.297 ms 81.019 ms 81.933 ms]
                        change: [-42.186% -41.637% -40.978%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 10 outliers among 100 measurements (10.00%)
  3 (3.00%) high mild
  7 (7.00%) high severe

Next steps would be to use this in more places (over current Iterator::next() approach), with accompanying benchmarks to prove its a performance improvement.

[WenyXu]

Thanks 🚀