Analysis of code on github shows that people use default parameters when they shouldn't. We can make that a little bit less bad by providing reasonable defaults. The default for n_estimators is not great imho and I think we should change it. I suggest 100.
We could probably run benchmarks with openml if we want to do something empirical, but I think anything is better than 10.

I'm not sure if I want to tag this 1.0 because really no-one should ever run a random forest with 10 trees imho and therefore deprecation of the current default will show people they have a bug.