[MRG] Reject complex input in soft-DTW barycenters

[mameikagou]

Fixes #167.

softdtw_barycenter can discard imaginary values during conversion and return a real-only result. Reject unsupported complex datasets and initial barycenters before conversion or optimization, including complex dtypes whose imaginary part is zero. Real-valued ragged inputs remain supported.

Validation: 21 barycenter tests pass, including NumPy object/ragged arrays and Torch complex and real inputs. The complex object-array regression fails on the base revision. git diff --check passes.

[codecov]

Codecov Report

❌ Patch coverage is 98.33333% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 94.33%. Comparing base (73ff407) to head (edb95fd).

Files with missing lines	Patch %	Lines
tslearn/barycenters/softdtw.py	95.23%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #720      +/-   ##
==========================================
- Coverage   95.26%   94.33%   -0.93%     
==========================================
  Files          83       83              
  Lines        7916     7976      +60     
==========================================
- Hits         7541     7524      -17     
- Misses        375      452      +77     

☔ View full report in Codecov by Harness.
📢 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.

[mameikagou]

All 24 Azure checks pass on edb95fd, and the patch coverage check passes at 98.33%.

The coverage job log reports a UCR/UEA dataset initialization error (_list_univariate_filename is missing); the dataset fixture then skips dependent tests. Comparing the final coverage tables with build 2164, 76 of the 77 additional uncovered lines are in unchanged dataset/test/utility files. The remaining line is the new fallback in softdtw.py.

No coverage thresholds or unrelated code were changed.