Erbs diffuse fraction model has discontinuities at transition points

[adriesse]

Is your feature request related to a problem? Please describe.
If you look closely at the piecewise definition of the Erbs diffuse fraction model you will see that it is discontinuous at the two transition points. This is not usually a problem, but when the Erbs model happens to be used in some kind of numerical optimization the discontinuities could cause problems.

Describe the solution you'd like
Add a new diffuse fraction function with a piecewise definition that closely matches the original Erbs model but is continuous. And while we're at it we can make the first derivative continuous too.

Describe alternatives you've considered
Modify the existing model function.

Additional context
This has been implemented and tested off line but I don't have the bandwidth for a PR currently. I thought I should mention it since it is somewhat relevant to #1753.