Create a general detrending method for mSSA

[The9Cat]

We currently have three very simple detrending schemes for mSSA:

1. The classic mean substracted, root variance divided scheme. This allows variations in any channel to be correlated with the same weight. Good for finding temporally covariant weak signals.
2. Mean subtracted, total variance divided scheme. This still puts each channel at the same overall level (by removing a DC component) but puts more weight on channels with larger variance.
3. Total power. This does not remove any constant floor (DC component) and therefore weights each channel by its gravitational energy for BFE coefficients or effective power for a general field.

Detrending can be more subtle and complex. For example, one might want to remove long-term trends by subtracting the fit to a smoothed low-order polynomial rather than the simple mean.

One general implementation might be a functional filter, passed to mSSA, which allows the user to specify the detrending and its inverse retrending as a simple class. pyEXP could provide some hardwired examples, but generally, the user would be able to create a derived trending class in Python which is passed to the mSSA code.

Other thoughts and comments welcome!