changes to ROMS reader for wetdry masks

[kthyng]

Main changes are to allow for the land_binary_mask as well as mask_rho, mask_u, and mask_v to have 3D or 2D dimensions since masks in ROMS can vary in time if they represented cells that can wet and dry in time. With the changes, if a variable wetdry_mask_rho is available, it will be used for the rho mask in place of mask_rho, otherwise will use mask_rho, and the same for the u and v masks. The masks for different grids use the same indices, same as before these changes, but now they are managed a little more simply in one location in the function. Additionally, more effort is made to match variables to the appropriate mask since sometimes they may not be a typical ROMS- output variable.

More changes:

• masks are saved in a dict in case they are reused across different variables
• the mask is applied to the variable array as a bool index directly to convert to nans. This is better than the previous approach of searching for the missing value and replacing that with nan because with wet/dry masks the dry cells are not always represented by a fill value and instead can only be found using the mask.
• added some more variables to check for whether to rotate or not and added log statements too.

[kthyng]

@knutfrode here is that PR I mentioned for dealing with wet/dry masks in ROMS (along a few other items).

[knutfrode]

Looks all good, and all tests and examples are passing, thus merging now. Thank you for the useful improvement.