Improve sink_rate argument in tracer_vertdiff()

[mnlevy1981]

tracer_vertdiff() has an optional argument for sink_rate ("rate at which the tracer sinks"), and @ElizabethYankovsky and I would like to use this feature for diurnal vertical migration of some MARBL tracers. We're running into two issues with the current implementation:

1. sink_rate is a scalar for each column, and we'd like it to vary with depth (model coordinate depth, not z)
2. As implemented, sink_rate must be non-negative but in some layers we want tracers to rise instead of sink

We discussed this at the last MOM6 telecon (Aug 14, 2023). I'm happy to work on this and eventually open a PR to the NCAR fork of MOM6; I think adding a depth dimension to the existing sink_rate parameter will be pretty trivial, and @Hallberg-NOAA made it sound like treating negative values as a rise rate is also feasible (but I'll ask for help when I get to that part).