add discrepancy function and related things

[petrelharp]

We would like to add to tsdate.evaluation:

1. break ties in match_node_ages by closest-in-time (and rename to match_nodes, which requires changing the code wherever the function is used)
2. add a discrepancy method that returns for the discrepancy between tsa and tsb, the tuple of
   a. the total shared span divided by total node span in tsa (which we need to compute); so this is proprotion of the span in tsa that is accurately represented in tsb; and
   b. the root-mean-squared discrepancy in time, with the average weighted by span in tsa.

[petrelharp]

@hfr1z3

[nspope]

match_node_ages isn't used anywhere else in this repo (only in my own evaluation/paper repo) and it's not documented-- so should be perfectly fine to just change the name in tsdate.evaluation

[hfr1tz3]

@petrelharp , I wrote up a discrepancy function with the desired outputs.
However, I need to double check if it correctly outputs the correct RMSE that @nspope is looking for.

I did run the following test case:
For simulated tree sequence ts, let sts = ts.simplify() and ets = sts.extend_edges().
Then tree_discrepancy(ets, ts) has discrepancy value of 1 and rmse of 0.
This makes sense as all of ets should be in ts, and there should therefore be zero error so I think it should be correct, but please double check if I missed something.

[petrelharp]

Okay, great. Let's see - this is the one on your fork, right? You may as well start a PR with these changes, that'll be easier to keep track of and discuss.

[petrelharp]

Hm, let's see. I agree with the check that you have. Here's some other possible checks:

1. Perturb the times of (some of) the nodes, so you know what the correct RMSE should be. (You could get a set of guaranteed legal perturbations by first iterating over the trees and finding an upper bound for each node by the lowest time of an ancestor across the trees; then only perturb nodes up.)
2. Make a very simple example where we know the right answer.