Velocity sometimes has NaNs in bottom level

[b-barton]

There are locations when there are NaNs in the velocity output by pyBDY in the bottom layer. The are not present everywhere. It may be something related to pybdy.utils.nemo_bdy_lib.rot_rep(), self.rot_dir, self.dst_gcos or self.dst_gsin containing NaNs.

[b-barton]

The nans come from nemo_bdy_extr_assist.interp_vertical(), sc_bdy_lev[ind_z] = np.nan because commenting this line removes the nans in this case. I suspect the problem stems from some mismatch in bathy on t grid and u/v grids.

[b-barton]

I've narrowed the problem down to nemo_bdy_zgrv2.get_bdy_depths(). bdy_H (the bathy) is sometimes less than bdy_z in layer -2. bdy_H is the np.ma.max(wz) and bdy_z is depths on t levels so maybe wz is being masked too shallow.
Since it is only happening to u and v it could be the same layer and i,j on the ut/vt grid is deeper than the tw grid for partial step. mbathy is used for masking and is on the t grid.

[b-barton]

I've found the problem. The vertical grid type was being misidentified as zps instead of sco. This meant pybdy was doing generating vertical uw, vw grids with w points at the minimum between the tw-grid points instead of average like with sco. This meant bathymetry max(w) could be shallower than the deepest unmasked t-level.

Wrongly identified zps grid where bottom t level (orange line) is in some places greater than bathy (blue line):

Correct sco grid with where bottom t level (orange line) is never greater than bathy (blue line):

[b-barton]

Added branch with changes that fix the bug.