Centralise data inputs to PModel in PModelEnvironment

[davidorme]

A set of linked changes to PModelEnvironment

• Following the proposed API sketch in PModel API for 2.0.0 #394, the suggestion for PModelEnvironment is that with the new PModel API all of the forcing variables can be set in the PModelEnvironment, so fapar and ppfd are set here. This makes it really easy to switch from standard to subdaily P Models.
• Free up the definition of extra variables to the PModelEnvironment. At the moment, downstream methods can require extra variables to calculate a method. The method registry setup makes this easily extendable, but at the moment these extra variables are hardcoded into the PModelEnvironment setup. We should change this to add any variables with the right dimensions via kwargs - this does lose specific variable bounds checking, but that is a relatively small price.
• Centralise calculation of tk in PModelEnvironment. Many of the calculations in the PModelEnvironment require scaling factors calculated using temperature in Kelvin. At the moment, all of these functions take temperature in °C and then convert internally. It would be much saner to calculate once when creating a PModelEnvironment instance and then store as an attribute and pass tk to functions not tc.