feat: complete/remove sci-kit learn wrapper to bms model predict method

[TheLemonPig]

Currently some of the experimentalists call the predict method of the bms model rather than the bms estimator itself. Because this seems to work for some users but not others AutoResearch/autora#224. Either way, the existence of the method provides a source of confusion, I think it is important to either remove it or provide it with the full sci-kit learn functionality of a predict method - being able to take both pandas and numpy inputs. Both changes would be simple. Reasons for and against it.

For:

• provides more ways to do things
• do not have to change the code for calling experimentalists

Against:

• confuses the user on how to use an estimator/theorist - they would not be able to call predict on regular sci-kit learn estimators in the same way

[TheLemonPig]

This was resolved by adding clarity in the documentation