Section - Efficiently working with large trees

[benjeffery]

For this section we wish to highlight how tskit allows efficient processing of large, single trees.

So far I have converted trees from http://hgdownload.soe.ucsc.edu/goldenPath/wuhCor1/UShER_SARS-CoV-2/ to tskit (thanks @jeromekelleher for starting code here) this gives a tree with 799318 nodes weighing in at 120MB (half of this is metadata).
map_mutations for each of the 27754 sites can be performed in an average of 19ms per site.

Next steps:

• Perf as a function of number of nodes/samples in the tree?
• Measure perf of subsetting operations based on metadata.
• Possible comparison with usher?

[jeromekelleher]

I'm imaging this section as a narrative showing that we can do real world things easily with tskit, efficiently, using the Python API. So, we say we loaded can load the trees into memory (x ms). Then identified identical samples. I guess a reasonable thing to aim for would be to duplicate some matUtils/usher operations using the Python API, and report the relative timings of these.