feat: Data history storage (for averaging)

[trisyoungs]

Follows #2204 and must be merged before #2203.

This PR introduces a new History template class to generalise (and simplify) the functionality of the old Averaging namespace. The idea is straightforward - introduce a sort of managed std::vector of data which can be easily formed into an average as well as serialised if required (e.g. as node data will require).

[rprospero]

Suggested change

	// Prune old data to get to the averagingLength
	while (history_.size() > averagingLength)
	history_.erase(history_.begin());
	// Perform averaging of the datasets that we have
	T averaged;
	auto weight = 1.0 / history_.size();
	for (auto &data : history_)
	averaged += *data * weight;
	// How many items to average
	auto length = history.size() < averagingLength ? history.size() : averagingLength;
	// Perform averaging of the datasets that we have
	T averaged;
	auto weight = 1.0 / length;
	for (auto &data : std::span(history_.rbegin(), ristory_.rbegin()+length))
	averaged += *data * weight;

This is an alternate implementation that doesn't lose history. This would allow us to look at multiple averaging lengths simultaneously (the current implementation essentially locks you into the shortest length). The disadvantage is that it can grow without bound, though that could be fixed by replacing the vector with a ring buffer.

This isn't a necessary change - just a suggestion.

[trisyoungs]

Interesting thought. While I like the idea, my concern is that this could add a lot of bloat as many of the objects being stored here are pretty chunky (I'm looking at you, PartialSet). As such, I would prefer to stick to the ultra-simple, shortest-length version for the time being.

[rprospero]

Looks good. Also gives us a nice framework for other stats (e.g. standard deviation, median, etc.)

One small suggested change, but, otherwise, things look good.