Fix rrule for * and add support for constant operations

[blegat]

I am a bit unsure what to do here. I think we should define rrule based on LinearAlgebra.dot so we should remove the definition of the rule for differentiate and keep the old one for *.

@harris-mit I would expect that the current rule for differentitation gives incorrect result when you use it with AD since it assumes a scalar product that is not what is used, does it work for you ?

[blegat]

@oxinabox this is the issue I mentioned to you at JuliaCon

[oxinabox]

@Sethaxen might have thoughts here