Replacing HGTConv with FastHGTConv

[puririshi98]

we have proven its correctness and speed, lmk if anything else needed to merge fasthgt->hgt

[codecov]

Codecov Report

Merging #7117 (ea043ee) into master (340d8b9) will decrease coverage by 0.03%.
The diff coverage is 98.90%.

@@            Coverage Diff             @@
##           master    #7117      +/-   ##
==========================================
- Coverage   91.30%   91.27%   -0.03%     
==========================================
  Files         437      436       -1     
  Lines       24002    23913      -89     
==========================================
- Hits        21915    21827      -88     
+ Misses       2087     2086       -1     

Impacted Files	Coverage Δ
torch_geometric/nn/conv/__init__.py	100.00% <ø> (ø)
torch_geometric/nn/conv/hetero_conv.py	98.52% <92.85%> (-1.48%)	⬇️
torch_geometric/nn/conv/hgt_conv.py	98.21% <100.00%> (-1.79%)	⬇️
torch_geometric/utils/hetero.py	64.28% <100.00%> (+6.07%)	⬆️

... and 1 file with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[wsad1]

Thanks for the update.
The changes look good to me. How did you access the correctness of fast_hgt_conv?

[puririshi98]

@rusty1s lmk if anything else needed to merge

[wsad1]

Thanks a lot.
Can we add an additional test where one of the node types is not a dst node type.
Currently all our tests are on the graph with node type author and paper and both author and paper are dst node types. We could add a new tests with node types author , paper , university and have university not be a dst node type. This will test L214 in hgt_conv much better.

[puririshi98]

thanks for approving @wsad1, i will merge this and add additional testing in a sep PR