Expose quantile regression objective and alpha in LightGbmRegressionTrainer (Resolves #7603)

[Sam7]

Description

This PR exposes the Alpha parameter in LightGbmRegressionTrainer.Options, bringing the ML.NET LightGBM wrapper into parity with the native C++ LightGBM library for Quantile Regression.

Resolves #7603

Background & Motivation

Currently, ML.NET users cannot perform native Quantile Regression (predicting the 5th or 95th percentiles instead of the mean) because the underlying LightGBM alpha parameter is hidden. This addition allows users to set Alpha directly in the options, unlocking probabilistic bounding and percentile estimation for regression tasks without needing to leave the native C# ML.NET ecosystem.

Changes Made

• Added public double? Alpha { get; set; } to LightGbmRegressionTrainer.Options (and mapped it to the LightGbmArguments).
• Ensured the parameter marshals correctly down to the native LightGBM C++ booster dictionary.
• Added/Updated unit tests to verify that Alpha is successfully passed through when specified.

API Impact & Safety

This is a strictly additive API change. The default behavior for standard L2 (Mean Squared Error) regression remains completely unaffected. If Alpha is not explicitly provided by the user, the trainer behaves exactly as it did before.

cc @michaelgsharp @luisquintanilla — Tagging you for visibility on the API addition! Since this exposes existing native C++ functionality and doesn't break default behavior, I'm hoping it's a smooth API review. Let me know if you need any adjustments to the tests or parameter mapping!

[Sam7]

@dotnet-policy-service agree

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 69.60%. Comparing base (9d809f1) to head (ff7b79c).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #7605   +/-   ##
=======================================
  Coverage   69.60%   69.60%           
=======================================
  Files        1484     1484           
  Lines      273604   273696   +92     
  Branches    27948    27951    +3     
=======================================
+ Hits       190433   190501   +68     
- Misses      75813    75837   +24     
  Partials     7358     7358           

Flag	Coverage Δ
Debug	69.60% <100.00%> (+<0.01%)	⬆️
production	63.84% <100.00%> (-0.01%)	⬇️
test	89.64% <100.00%> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
...Microsoft.ML.LightGbm/LightGbmRegressionTrainer.cs	80.80% <100.00%> (+2.22%)	⬆️
...osoft.ML.Tests/TrainerEstimators/TreeEstimators.cs	98.06% <100.00%> (+0.21%)	⬆️

... and 6 files with indirect coverage changes

🚀 New features to boost your workflow:

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

[Sam7]

The failing CI appears to be unrelated to this PR. The error in Helix is Microsoft.ML.Tests.TrainerEstimators.TrainerEstimators.SdcaLogisticRegression on osx.15.arm64.open, whereas this PR only changes the LightGBM regression trainer/options/tests.
(I don't seem to have permission to rerun failed jobs in ADO)

Happy to make any adjustments if needed, but this looks like an unrelated test failure or flaky CI issue.