Low level pair coalescence counts

[nspope]

Low level extension of #2915

[codecov[bot]]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 89.62%. Comparing base (e6483fc) to head (62069ec).

Additional details and impacted files

```diff @@ Coverage Diff @@ ## main #2932 +/- ## ======================================= Coverage 89.62% 89.62% ======================================= Files 29 29 Lines 30176 30176 Branches 5874 5874 ======================================= Hits 27044 27044 Misses 1793 1793 Partials 1339 1339 ``` | [Flag](https://app.codecov.io/gh/tskit-dev/tskit/pull/2932/flags?src=pr&el=flags&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=tskit-dev) | Coverage Δ | | |---|---|---| | [c-tests](https://app.codecov.io/gh/tskit-dev/tskit/pull/2932/flags?src=pr&el=flag&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=tskit-dev) | `86.21% <ø> (ø)` | | | [lwt-tests](https://app.codecov.io/gh/tskit-dev/tskit/pull/2932/flags?src=pr&el=flag&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=tskit-dev) | `80.78% <ø> (ø)` | | | [python-c-tests](https://app.codecov.io/gh/tskit-dev/tskit/pull/2932/flags?src=pr&el=flag&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=tskit-dev) | `88.72% <ø> (ø)` | | | [python-tests](https://app.codecov.io/gh/tskit-dev/tskit/pull/2932/flags?src=pr&el=flag&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=tskit-dev) | `98.97% <ø> (ø)` | | Flags with carried forward coverage won't be shown. [Click here](https://docs.codecov.io/docs/carryforward-flags?utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=tskit-dev#carryforward-flags-in-the-pull-request-comment) to find out more.

[nspope]

I'd like to generalize this algorithm slightly:

• Currently, the output is either a num_windows by num_nodes array (which is very large), or a num_windows by num_time_windows array where the counts are summed within time windows.
• Conceptually, the "nodes" output gives the empirical distribution of pair coalescence times in windows across the genome. That is, for each window we have a vector of RVs (node times) and a vector of weights (pair coalescence counts).
• From this distributional viewpoint, there's lots of useful things that may be calculated: the empirical CDF, quantiles, moments, coalescence rates, etc. (of which the "sum in time windows" option of the current implementation is one special case)

So, I think it'd be useful to take the current algorithm, and have it apply a summary function at the end of each window. This would let us calculate any useful summary statistic without having to create a potentially humongous array (windows by nodes by indexes) as an intermediate.

The API would stay the same-- later on, we could add named methods for various summary statistics, and potentially eventually expose a "general summary stat" interface, like is done for the other statistics.