[21pt] Incorporate ML roughness into SRCs

[CarsonPruitt-NOAA]

Thoughts from Fernando: FIM v4.X Eval: When testing the ML derived Manning's N with FIM v4.X, I think we should make sure we're using it with the default FIM v4.X SRCs adjusted to include the ML derived 3d channels without any SRC calibration. I would then evaluate the output with this first, calibrate again and then evaluate the calibrated results to compare before and after as well as the FIM v4.X results without any ML enhancements. FIM v4.X GMS: When testing FIM v4.X mapping, make sure to adjust the SRCs for all FIM v4.X branches and then test the mosaiced FIM v4.X mapping output instead of just branch 0. This approach would best evaluate the impact to FIM v4.X mapping. We already know the branch 0 significantly underpredicts which is why we don't use it alone in real-time. Our target is to understand the impact on the FIM capabilities we use in the real-time services.