feat: kornia binary morph ops

[nkemnitz]

No description provided.

[nkemnitz]

@supersergiy Do you want tests here or is that more of a delegation thing?
Also, any reason to keep the custom binary_closing around?

[codecov]

Codecov Report

Merging #428 (8b6a13f) into main (0525c92) will not change coverage.
The diff coverage is 100.00%.

@@            Coverage Diff            @@
##              main      #428   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files          120       120           
  Lines         3588      3618   +30     
=========================================
+ Hits          3588      3618   +30     

Files Changed	Coverage Δ
zetta_utils/tensor_ops/mask.py	100.00% <100.00%> (ø)

[supersergiy]

This looks really cool!

Although the new functions don't have if/for statements, they're reasonably complex and would benefit from testing. The upside is that because there's no if's, each function will need only a single test case to be fully covered.

Re deprecation:
The kornia_closing requires kernel to be passed in explicitly, while the "custom" binary_closing just assumes something reasonable. The explicit kernel is useful in some scenarios, but making it a required argument will make it a hassle a lot of the time. I think ideally we'd make the kernel an optional argument, which would let us deprecate the non-kornia alternatives.

Also, the kernerl argument should probably be typed, unless there's something I'm missing.

[nkemnitz]

Set default kernel to square footprint of width=3, removed unnecessary conversion to floating point, added documentation

[supersergiy]

Looks good! It's great that the kernel typing is more precise now - Tensor instead of TensorTypeVar. Now that the new kornia based operations have width, should we deprecate the old coarsen and dilation too?