fix(python): Preserve Duration type in timedelta multiplication

[Zakir032002]

Handle timedelta multiplication in both mul and rmul to preserve
Duration type instead of silently casting to float. This fixes
datetime/float encoding round-trips commonly used in time-series analysis.

Changes:

• Added explicit timedelta handling in Series.mul() before the
  _arithmetic fallback
• Added explicit timedelta handling in Series.rmul() for reverse
  multiplication
• Both paths use F.lit(timedelta) to preserve temporal type information
  via the expression engine

Fixes #26205

1. I used AI assistance to help analyze the root cause .
2. I have reviewed all changes myself and confirm they are correct,
   minimal, and consistent with the existing codebase.

[codecov]

Codecov Report

❌ Patch coverage is 0% with 4 lines in your changes missing coverage. Please review.
✅ Project coverage is 81.00%. Comparing base (32dbe12) to head (74cc774).
⚠️ Report is 6 commits behind head on main.

Files with missing lines	Patch %	Lines
py-polars/src/polars/series/series.py	0.00%	2 Missing and 2 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #26438      +/-   ##
==========================================
- Coverage   81.02%   81.00%   -0.02%     
==========================================
  Files        1782     1782              
  Lines      243116   243120       +4     
  Branches     3078     3080       +2     
==========================================
- Hits       196980   196940      -40     
- Misses      45333    45375      +42     
- Partials      803      805       +2     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[orlp]

I don't think this should be fixed by adding a hack in the Python code. Can you trace where the current multiplication implementation goes and fix it there?