perf: Reduce memcopy in parquet

[nameexhaustion]

It should be sound since it matches what we did a long time ago

polars/crates/polars-io/src/parquet/read/read_impl.rs

Line 496 in 04714b9

Ok(mmap::ColumnStore::Local(self.0.deref()))

[codecov]

Codecov Report

Attention: Patch coverage is 94.73684% with 1 line in your changes missing coverage. Please review.

Project coverage is 80.21%. Comparing base (8a76dad) to head (45f1690).
Report is 70 commits behind head on main.

Files with missing lines	Patch %	Lines
crates/polars-io/src/mmap.rs	75.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #19350      +/-   ##
==========================================
- Coverage   80.24%   80.21%   -0.04%     
==========================================
  Files        1523     1523              
  Lines      209545   209790     +245     
  Branches     2434     2434              
==========================================
+ Hits       168148   168281     +133     
- Misses      40842    40954     +112     
  Partials      555      555              

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

[ritchie46]

This isn't safe if the self is dropped before the MemSlice.

I see that MemSlice has an inner that can have a bytes::Bytes variant. If we have an ReaderBytes::Owned we can move that vec into the MemSlice.

I think we should also change the type form Vec<u8> to Bytes so we can freely clone it.