Timing fix for high-latency line subtraction

This MR fixes a minor issue that led to at least one instance of inadequate line subtraction in the generation of C01 data. Updates to the length of the running median of the line subtraction TFs were not being applied at the optimal times. The error was that updates were timed based on the timestamps of input data to the running median, rather than the timestamps of the output data, as was originally intended. Here is an example spectrogram of the NOLINES channel that shows the problem that this addresses:

After the fix, the subtraction looks better at this time: