Add option Deterministic to LightGbmBinaryTrainer.Options and LightGbmMulticlassTrainer.Options

[grzskoBgs]

Is your feature request related to a problem? Please describe.
LightGBM is known to give non-deterministic results when run with more than one thread. It affects the reproducibility of results. With version 3.3.5 of LightGBM a new parameter "deterministic" was introduced: https://lightgbm.readthedocs.io/en/v3.3.5/Parameters.html#deterministic. However, I don't see this option as available for LightGbmBinaryTrainer.Options in the latest version of ML.NET (4.0.0).

Describe the solution you'd like
The property "Deterministic" is available for LightGbmBinaryTrainer.Options and LightGbmMulticlassTrainer.Options with the same interpretation as the "deterministic" parameter of the Python LightGBM implementation.

Is it possible, what is the expected resolution time?

Describe alternatives you've considered
We were considering running LightGbmBinaryTrainer with option NumberOfThreads equal 1. However, it significantly slows down the execution time. We want to consider a built-in solution for assessing the reproducibility of results.

Additional context
No

[michaelgsharp]

Good idea. We'll get this added.