Add linear regression model

[christophmark]

Observation model like the AR1, but with a constant instead of the auto-correlation term. This allows to independent Gaussian noise with non-zero mean, or, in the case that first order data differences of data points are provided, basic linear regression (with static transition model). Further, this approach also allows to identify trends dynamically (with other transition models). Especially the case of abrupt on/off-sets (regime-switching process) of trends might yield new insights into some systems...

[christophmark]

Is now implemented indirectly by introducing a Gaussian observation model. When fed with first order data differences, this allows to employ linear regression.