Compatibility fix for XGBoost 3.2

[hcho3]

Since dmlc/xgboost#11814, XGBoost stores leaf weights as a separate field for multi-target trees.

Since multi-target trees are an experimental feature, we do not provide a compatibility layer for older versions of multi-target trees.

[hcho3]

@trivialfis Can you help review? This updates Treelite to support new layout for multi-target trees.

[codecov]

Codecov Report

❌ Patch coverage is 87.50000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 84.38%. Comparing base (411113b) to head (5c785fd).
⚠️ Report is 1 commits behind head on mainline.

Files with missing lines	Patch %	Lines
...el_loader/detail/xgboost_json/delegated_handler.cc	87.50%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##           mainline     #655      +/-   ##
============================================
+ Coverage     84.35%   84.38%   +0.03%     
============================================
  Files            75       75              
  Lines          6653     6653              
  Branches        543      543              
============================================
+ Hits           5612     5614       +2     
+ Misses         1041     1039       -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.

[trivialfis]

The change looks good. What do you think of the new layout? Theoretically, the leaf weight can be omitted if reduced gradient is not used. But making an optional field makes the code and more complex.

[hcho3]

@trivialfis Is reduced gradient only applicable for multi-target trees? If so, I like the layout.

[trivialfis]

Yes, vector leaf only