SAMOS: Add constant extrapolation option

[brhooper]

Addresses https://github.com/metoppv/mo-blue-team/issues/1051

This PR adds the capability to use constant extrapolation to the GAMPredict class and the relevant SAMOS classes and CLIs.

When extrapolating along a single term in a GAM, the default behaviour is to extend the trend of that term when predicting for input values which are outside the bounds of the training data. When using constant extrapolation, this will instead treat any values outside of the training range as equal to the smallest/largest value in the training range, as required.

For example, suppose our fitted GAM term were a simple linear term y = x + 2 trained on x values in the range [0, 10]. Predictions from this term would look as below, using the two different extrapolation approaches:

	Default extrapolation	Constant extrapolation
x	2	2
y	4	4

	Default extrapolation	Constant extrapolation
x	-1	-1
y	1	2

	Default extrapolation	Constant extrapolation
x	100	100
y	102	12

Testing:

• Ran tests and they passed OK
• Added new tests for the new feature(s)

[bayliffe]

Everything passes and I cannot see any issues with this. Good work.

[gavinevans]

Thanks @brhooper 👍

I think these changes are fine.