Revisit Fringe-Rate Filter Design Algorithm

[r-pascua]

@steven-murray found that we were having some issues generating fringe-rate filters for long baselines at high frequencies when using the "deinterleaved" time series. The issue boils down to alaising of the "main lobe" fringe rates, as we can see in this plot of the fringe-rate profile for a baseline with >200 m east-west projected length in the highest spectral window:

This causes the cumulative power to take on an odd shape, as shown below:

Since the fringe-rate filter bounds are determined from the cumulative power, the fringe-rate profile aliasing generates bogus FRF bounds. I will make a PR soon with a quick fix to this, but I wanted to make a note to remember to revisit this later to see if there's a better solution than the hacky one I came up with.