Fractional polariation error bugs and other things from Shinsuke about rm-tools HOT 1 CLOSED

I looked into this more carefully today. Shinsuke was completely right: the fractional Stokes errors were not being computed properly, and were not used properly through the rest of the program. I thought about it a bit, and there was also a somewhat deeper problem: the equation in the fractional Stokes errors was wrong. It assumed that the error in Stokes I was the channel error, but we're actually using a model of Stokes I, which presumably has much smaller errors.

I've changed it to ignore errors in Stokes I, and just normalize the Stokes Q/U errors by the Stokes I model. Quantifying the errors in the model will be difficult, and should be a very minor contribution for all sources bright enough to have potential detections in polarization. So the fractional Stokes errors will be slightly smaller than before.

I've also changed how the errors propagate through. Before, the actual Stokes Q/U errors were used, but the fractional Stokes were used in the RM synthesis, so there was a contradiction. I've changed it over to using the fractional errors to determine the weights and in the estimation of dFDF. This will down-weight channels with fainter Stokes I, as should be the case.

The changes in the error calculations ripple down through all the output values. For flat spectra nothing will change, but for steep spectra the output values will be a bit different, probably for the better.

from rm-tools.