Automatic sampling & transform cache disabling

[George3d6]

We need to:

a) Start sampling training data when the dataset is too large (e.g. can barely fit in RAM)
b) Automatically disable the transformed cache for lightwood when the dataset is too large (e.g. ~5-10% or more of the total RAM)

For now let's say we have 4 potential state for the data:

• small
• medium
• big
• huge

Determined as a how much of the available RAM the pandas dataframe uses.

If small --- do nothing
If medium --- use the current sampling algorithm to sample the data analysis data
If big --- same as medium + set transform cache disable in the lightwood configuration for training
If huge --- same as big + use the current sampling algorithm to sample data for the testing, training and validation sets.