perf(rust): faster decode on Parquet HybridRLE

[coastalwhite]

This PR performs several optimization surrounding the HybridRLE decoding:

• A lot of iterators in favor of collecting and buffering
• Flatten nested iterators into a single iterators
• Remove many memcopies in bitpacked::decode.

Overall, my microbenchmarks showed an improvement of roughly 10x for the HybridRLE decoding. In order to get a better view on how to much impact this has on totality of parquet decoding I did a benchmark using the Yellow NYC Taxi dataset. This got me a roughly 1.8x speed-up for general parquet decoding.

[codecov]

Codecov Report

Attention: Patch coverage is 92.30769% with 20 lines in your changes missing coverage. Please review.

Project coverage is 80.85%. Comparing base (7a1be3b) to head (93a6ee6).
Report is 11 commits behind head on main.

❗ Current head 93a6ee6 differs from pull request most recent head 882e362

Please upload reports for the commit 882e362 to get more accurate results.

Files	Patch %	Lines
...s-parquet/src/parquet/encoding/bitpacked/decode.rs	89.77%	9 Missing ⚠️
...ars-parquet/src/parquet/encoding/hybrid_rle/mod.rs	97.01%	4 Missing ⚠️
...quet/src/arrow/read/deserialize/binary/decoders.rs	25.00%	3 Missing ⚠️
...rquet/src/arrow/read/deserialize/dictionary/mod.rs	50.00%	2 Missing ⚠️
.../arrow/read/deserialize/fixed_size_binary/basic.rs	0.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #17208      +/-   ##
==========================================
+ Coverage   80.84%   80.85%   +0.01%     
==========================================
  Files        1464     1465       +1     
  Lines      192016   192190     +174     
  Branches     2743     2745       +2     
==========================================
+ Hits       155243   155404     +161     
- Misses      36262    36281      +19     
+ Partials      511      505       -6     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.