Rethink consistency between latitudinal gradient, seasonality and monthly means

[znichollscr]

As discovered in #71, there are times where our current techniques lead to latitudinal gradient and seasonality values which are bigger than the global-mean, which leads to negative values in the output. The solution in #71 is a workaround, shrinking the latitudinal gradient and seasonality until we don't get negative values.

A better solution would be to re-think how we calculate the latitudinal gradient and seasonality, to ensure that this sort of difference can't occur.