perf(python): Avoid needless copy when converting chunked Series to NumPy

[stinodego]

Ref #14334

We used to always rechunk first when converting to NumPy. This is not great because:

• If we go through the copy path (e.g. Series contains nulls), we iterate over the values anyway and rechunking is unnecessary.
• If we go through the 'view' path (e.g. numeric Series without nulls), we do a copy (rechunk) but we end up with a non-writable array. Better to go through the copy path directly to ensure a writable array. Otherwise making the array writable is an additional copy.

Also changed some into_iter calls to iter.

[codecov]

Codecov Report

Attention: Patch coverage is 97.61905% with 1 lines in your changes are missing coverage. Please review.

Project coverage is 80.98%. Comparing base (3892c75) to head (80a3d5d).

Files	Patch %	Lines
py-polars/polars/series/series.py	75.00%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #16178      +/-   ##
==========================================
- Coverage   80.99%   80.98%   -0.02%     
==========================================
  Files        1387     1387              
  Lines      178832   178828       -4     
  Branches     2877     2876       -1     
==========================================
- Hits       144844   144821      -23     
- Misses      33496    33513      +17     
- Partials      492      494       +2     

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