Revisit bottom roughness generation (#105)

 - Compute the buoyancy frequency squared (N^2) from WOA T/S data.
 - From N2, computes the mode-1 internal tide wavelength (lambda1) at a given tidal frequency (e.g. M2)
 - Smooth high-resolution bathymetry (SYNBATH) using a polar, Gaussian-weighted average with radius lambda1, producing a physically meaningful mean depth.
 - Compute depth variance relative to this smoothed depth, getting bottom roughness independent of model grid resolution.
 - Interpolate the resulting roughness field onto model grids using esmf regridding. For low resolution configs, conservative_normed is used, while for moderate to high resolution configs, bilinear interpolation will be used.
 - Complete metadata and provenance.
---------

Co-authored-by: Andrew Kiss <31054815+aekiss@users.noreply.github.com>

Author: minghangli-uni