Couple adaptive ocean

[JanJereczek]

This PR is a substantial refactoring of the code:

• Current and reference states are now abbreviated in fip::FastIsoProblem by fip.now and fip.ref
• The load is only applied for a given mask if sea level is interactive
• Code is not only relative anymore, i.e. it does not only consider anomalies. Instead, the update of the bathymetry and the columns is now embedded
• The gravity perturbation of the sea-surface height takes into account the fact that the barystatic sea level has changed
• Option to ignore lithospheric gradients has been deleted
• PREM and elastic Green coefficients have been moved to dataloaders.jl
• The functions exported were reduced
• Docstrings improved for some functions
• Piecewise linear approx of nonlinear ocean surface has backup method = piecewise constant
• External forcing of the (barystatic) sea level is now possible
• Memory leaks fixed - largely by making FFT more type specific and dropping the idea of user-defined BC.
• Draft of Test3B - benchmarking against Abaqus/ASPECT
• support vector for analytic solution is not a user input anymore

[JanJereczek]

Additionally, further changes have been made:

• InplaceConvolution is slower than its equivalent in DSP.jl but barely assigns memory. This is what we want since the operation is fast anyway but we hit a memory limit when using conv. Memory allocation low enough to rune glacial cycle on 350*350 grid ( even for interactive sea level). Therefore, DSP.jlremoved.
• Readme updated with video of ICEW6G glacial cycle
• Deleted anything related to time-variable viscosity
• column_anom is a single function that is easier to maintain than the previous ones
• blurring utils for masking
• output can be either sparse or dense
• generic way to apply BCs without requiring allowscalar()