Benchmark Hummingbird for model compilation

Benchmark Hummingbird: https://github.com/microsoft/hummingbird

We should test on XGBoost & LightGBM, and try out the GPU support, to see if we can get a meaningful inference speedup.

Start by reading the Hummingbird project at https://github.com/microsoft/hummingbird and identify how to benchmark XGBoost and LightGBM inference with GPU support. Run comparable benchmarks and document whether they produce a meaningful inference speedup; the issue does not name repository files, tests, or a specific success threshold.

Written by the indexing model from the issue text.