feat: framework for parallelizable statistic computation

[laggycomputer]

• Introduce the SiteComposable trait, guaranteeing that a GlobalStatistic can be computed from its parts indepedently, concurrently, and potentially out-of-order.
• Use this trait to create generic code capable of parallelizing any such statistic. There is an example instead.

[molpopgen]

It seems that we need some tests? A computation done in parallel should be "close enough" to the serial version?

[laggycomputer]

We definitely need tests, and we should at least prepare, if not also provide, a sketch demonstrating the use of indexed queries to do parallelism without the bottleneck of writing and cloning a records on a single thread

[laggycomputer]

As discussed today:

• Let's sanity check at least one other stat.
• We should combine components in an order besides the conventional left-to-right fold, to verify that this works as intended.

[molpopgen]

@laggycomputer -- is this ready for more look/review?

[laggycomputer]

We're still missing one other stat, but feel free to look at the test for pi to see if it makes sense

[molpopgen]

I think that the strategy looks sound. Boy, it would be handy to abstract this out for use for other stats. (Can definitely not do that for this PR though!)

[molpopgen]

Approve. Only issue (which we can address later) is if we want anything "parallel" to be feature-gated? I'm agnostic on that at the moment.