Fix handling of global bias for binary:logitraw objective of XGBoost

[hcho3]

Prior to dmlc/xgboost#6517, XGBoost was incorrectly applying the inverse sigmoid function the global bias when the objective was set to binary:logitraw. (The correct action is to apply no transformation.) Treelite was mimicking XGBoost's behavior.

XGBoost 1.3.1 and later has dmlc/xgboost#6517 and thus will correctly handle the global bias, so update Treelite as well.

[codecov-io]

Codecov Report

Merging #242 (2d0549b) into mainline (191a474) will increase coverage by 65.78%.
The diff coverage is 100.00%.

@@               Coverage Diff               @@
##             mainline     #242       +/-   ##
===============================================
+ Coverage       17.93%   83.72%   +65.78%     
  Complexity         44       44               
===============================================
  Files              65       93       +28     
  Lines            4555     6925     +2370     
  Branches           42       42               
===============================================
+ Hits              817     5798     +4981     
+ Misses           3714     1103     -2611     
  Partials           24       24               

Impacted Files	Coverage Δ	Complexity Δ
src/frontend/xgboost/xgboost.h	100.00% <ø> (+100.00%)	0.00 <0.00> (ø)
src/frontend/xgboost.cc	79.70% <100.00%> (+79.70%)	0.00 <0.00> (ø)
src/frontend/xgboost_json.cc	69.60% <100.00%> (+69.60%)	0.00 <0.00> (ø)
src/frontend/xgboost_util.cc	97.43% <100.00%> (+97.43%)	0.00 <0.00> (ø)
src/predictor/thread_pool/spsc_queue.h	93.87% <0.00%> (-0.72%)	0.00% <0.00%> (ø%)
src/logging.cc	100.00% <0.00%> (ø)	0.00% <0.00%> (ø%)
src/typeinfo.cc	100.00% <0.00%> (ø)	0.00% <0.00%> (ø%)
python/treelite/sklearn/__init__.py	90.90% <0.00%> (ø)	0.00% <0.00%> (?%)
python/treelite/sklearn/gbm_multi_classifier.py	94.11% <0.00%> (ø)	0.00% <0.00%> (?%)
python/treelite/contrib/gcc.py	100.00% <0.00%> (ø)	0.00% <0.00%> (?%)
... and 74 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 191a474...2d0549b. Read the comment docs.