Update Dask RF to use the new distributed algo

[chyunsu3]

• Wire Dask RF training to RAFT communicators
• Retain one fitted worker model as the canonical model for prediction and get_combined_model().
• Remove obsolete components: per-worker n_estimator, ignore_empty_partitions, partial inference, etc
• Update terminology: FIL -> nvForest
• Use global n_rows when constraining certain hyperparameters
• Update pytests

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[RAMitchell]

Cleaned up code looks great!

We shouldn't outright remove parameters. Instead warn the parameter is no longer used and flag deprection.

I think only 1 stream ever gets used in the distributed version.

comms.init() is outside the try/finally

[chyunsu3]

only 1 stream ever gets used in the distributed version.

Should the distribute algo use multiple streams? The distributed algo uses a single stream only. Perhaps it's a good idea to throw a warning when n_streams>1

[RAMitchell]

@chyunsu3 yes the parameter is currently unused.