Time series: support max_delay and min_delay per-column

It would be great if we could override the dataset-wide setting of max_delay and min_delay.

Example 1: compute windowed average of "feature A" with window size of 7. Compute windowed average of "feature B" with window size of 30.

Example 2: compute "feature A" delayed from 4 to 10 time units. Compute "feature B" delayed from 4 to 20 time units.

Note the impl will probably have to take the minimum min_delay and maximum max_delay in order to determine the cutoffs for shifting the data around.