Depreciate pangolearn

[rmcolq]

This seemed simpler than I was expecting given the conversations I had with Áine (not tagged deliberately). The intention here is:

1. If pangolearn or fast mode specified, prints a warning and continues with usher mode instead
2. If a datadir provided and pangolearn specified, allows pangolearn mode (for backwards compatibility) NB automatically assumes this is an old data dir so versioning will go backwards
3. The next data release will remove the randomforest files from pangolin-data so that only the up to date models are there.

This all seemed to work when I tested it locally, but would be great to have another set of eyes on it.

[AngieHinrichs]

Thanks so much @rmcolq, this looks great! I pulled down the changes and did some testing too, and it all behaved as expected.

I pushed one little change to remove a debugging print statement.

As far as I'm concerned this is fine to merge into the master branch any time. Before we tag the next (pre)release on the repo, I want to make one more little change for v4.3: I want to remove my max_count/max_lineage logic in usher_parsing in report_collation.py that overrides usher's selected lineage with the lineage that gets the highest number of equally parsimonious placements. That's been a bad idea at least since Omicron, maybe earlier. So after this is merged, I'll make that little tweak.

[Looks like I also forgot to update pangolin/data/data_compatibility.csv to include v1.19. And going forward (for pangolin-data v1.20 & beyond) the minimum version in data_compatibility.csv should probably be 4.3 instead of 4.]