Add H2O cluster memory best practices for using XGBoost
We don't have any "pro-tips" about how to allocate memory to the H2O cluster when using XGBoost. When running XGBoost, or AutoML (which includes XGBoost), there are different suggested settings for how much memory to allocate to the H2O cluster. Usually you can get away with allocating most of the free memory for H2O, however, XGBoost needs its own memory, to be very conservative, we recommend that you allow half of whatever free memory you have to H2O and the other half leave free so that XGBoost can consume it.
Example: If you have 102G RAM on your machine, then (at most) set `max_mem_size` in H2O to 50G, so that XGBoost can have another ~50G for itself.
We should add a section to both the XGBoost and AutoML user guide pages about this. It would probably also be useful to add a note about this in the Starting H2O section (either linking to the XGBoost page or duplicate the info).