perf: Use two-pass algorithm for csv to ensure correctness and SIMDize more ~17%

[ritchie46]

fixes #19078

Fixed so that we always have chunks that are valid. As we now have a two pass algorithm, made sure to SIMDize more. What is lost by the two pass, is gained by SIMD. On NY city trip data I have ~17% perf improvement.

The good thing is that we are now always fast, as we don't have to fallback to single threaded code anymore when we encounter a complicated/quoted file.

[codecov]

Codecov Report

Attention: Patch coverage is 98.65772% with 6 lines in your changes missing coverage. Please review.

Project coverage is 79.78%. Comparing base (310b482) to head (1a1d3ab).
Report is 8 commits behind head on main.

Files with missing lines	Patch %	Lines
crates/polars-io/src/csv/read/parser.rs	97.77%	4 Missing ⚠️
crates/polars-io/src/csv/read/read_impl.rs	99.29%	1 Missing ⚠️
crates/polars-io/src/csv/read/splitfields.rs	99.00%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main   #19088   +/-   ##
=======================================
  Coverage   79.78%   79.78%           
=======================================
  Files        1531     1531           
  Lines      208568   208561    -7     
  Branches     2913     2913           
=======================================
+ Hits       166405   166409    +4     
+ Misses      41612    41601   -11     
  Partials      551      551           

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

[tooptoop4]

i find v1.10 is slower (2.6s vs 1.7s) than v1.7.1 for my 6gb quoted file for doing sum of a column