Add constraint that each uncertainty variable should be linked to a single observation variable

[shunt16]

From Zhav Loizeau in #1:

Should there be a constraint that each uncertainty variable should be linked to a single observation variable?

[shunt16]

Are there cases where it would be safer to have this?

On the other hand, are there cases where we might not want to allow this?

[EmmaWoolliams]

It may be the case that more than one input quantity has exactly the same uncertainty (e.g. the Earth count might have the same distribution as the onboard calibration target measured count) ... but we really don't want to encourage that short hand because it stops people coming back to it later when they realise they aren't the same ... and it makes them confuse common errors and common uncertainties - that's why the uncertainty tree diagram always had the same branch duplicated.

[EmmaWoolliams]

This is linked to questions about whether we include sensitivity coefficients, which I mentioned in #7 (comment)

[EmmaWoolliams]

Do we want to consider the case where there may be a correlation between two measured values, e.g. if they are both determined through the same fitting routine (e.g. significant wave height and sea level anomaly in radar altimetry of the ocean surface), or if they are the real and imaginary parts of something (#4 (comment)), and so on.