- each crop model must implement types `CropType <: AbstractCropType`
  and `CropState <: AbstractCropState`

- `AgricultureModel` and `FarmPlot` are not parametrised structs
  anymore to allow for different crop models active in one simulation

- no majority counting of Pixels implemented yet, soil type is
  currently still determined by the first pixel of a FarmPlot

The data/crops/almass/crop_data_general.csv file now stores a column
`phase_before_harvest` to indicate in which growing phase the plant
can be harvested.  This is `marchfirst` for most crops, and `sow` for
maize.

In theory, with the current setup there could be a problem for
peas/beans if they are planted after March, 1 (the planting dates are
15 February to 15 March).  Planted before March 1st, the last growth
phase before harvest would be `marchfirst`.  But if planted after
March 1st, the last growth phase before harvest would be `sow`.  The
GDD curves for this crop only exist for `sow`, so probably planting
before March 1st should not be allowed for this crop.

Currently the sowingdensity column is still treated as type `String`,
as there is an entry of "35/80".  Once this entry is clarified, the
type should probably be `Float64`.

This reverts commit b077630d.

`sow!(::ALMaSS.CropState, ...)` calls `setphase!`, which needs to know
which growth curve to use.  Some crop types (e.g. "permanent
set-aside") do not have a lownutrientgrowthcurve, which then causes an
error when `setphase!` tries to access the `missing` growth curve.

The highnutrientgrowthcurve would only be chosen when the cropstate
had `fertilising` in its events.  As the events for a cropstate are
now `empty!`ied in `sow!`, it's not possible to set the `fertilising`
event before calling `sow!`.

This previously didn't cause a bug, as the logic for which growth
curve to choose was the wrong way around, choosing the
highnutrientgrowth curve when no `fertilising` event was set for the
cropstate.

Previously, we chose the lownutrientgrowth curve when the was a
`fertiliser` event for the cropstate, otherwise the highnutrientgrowth
curve.

It should be the other way around.

The code seems to work for crop plants but currently doesn't yet work
for non-crop plants such as "permanent set-aside", which grow to
enourmous height and do not get reset at the end of the year.  I think
this is due to different treatment of the growth phase transitions for
these plants.

closes #109