Add calculation of minimum spanning trees

[nickjcroucher]

This is an inadequate implementation currrently. A design choice needs to be made about whether:

• multiple trees should be returned if a network is not fully connected (I am not aware of a viewer of multiple MSTs since eBURST v3, so this would limit visualisation possibilities)
• fully-connected networks could be constructed (but this is too slow for large datasets)
• a threshold could be identified for making the network sparse but fully connected (this seems to require constructing an MST)
• connections could be made between distinct components(either arbitrary or calculated between selected references - these could be the "founders", as identified by eBURST CCs, for which there is already code when selecting the dendropy seed_node)

[johnlees]

@nickjcroucher Could you merge the master branch in?
Then I should be able to take a look on Monday

[nickjcroucher]

Will do - just for inspiration, here is the MSTree code from Grapetree, which may be helpful: https://github.com/achtman-lab/GrapeTree/blob/master/module/MSTrees.py

[johnlees]

Also I'm not sure why the tests aren't running on this PR, I can try and sort that out too

[johnlees]

Also I'm not sure why the tests aren't running on this PR, I can try and sort that out too

I've edited the config slightly on master, so hopefully merging master in should get the testing running on this branch

[johnlees]

Looks good overall! Changes here are mostly minor formatting/style things.

Still needed before merge:

• Correct implementation of sparse matrix joining
• Pin the version of dendropy in environment.yml now the midpoint rooting is fixed (and double-check NJ trees are correct, if we haven't already)
• Add a test which runs the MST
• Add an example to the visualisation docs
• Add cuGraph version (see also GPU accelerated graphs #87)
• Draw the MST as a network rather than tree (cytoscape or graphtool)
• Alternative way of sparsifying graph (threshold/max memory approach)

I will raise a separate issue with ideas which might improve this, and we can decide whether to add this in this merge, or work on them more later