Source rejection by domain (cell, material, universe)

[paulromano]

This PR implements the ability to specify a set of a cells, materials, or universes that are used to "filter" sampled source sites and closes #1829. The way it works is:

1. The user specifies Source(..., domains=[...]), where domains is a list of either cells, materials, or universes that will be used for rejection.
2. At runtime, positions will be sampled from whatever spatial distribution was specified (e.g., openmc.stats.Box), but then those points will be checked against the list of domains, if present. If a point is not within the specified list of domains, it will be rejected.

The attribute handling for the Source class follows the same design as for the VolumeCalculation class. Namely, even though the constructor specifies domains, this is turned into domain_type and domain_ids attributes. This is done so that a Source object doesn't have to be inextricably linked to actual geometry objects.

@pshriwise and I discussed other overall design alternatives for this feature. In particular, we discussed the possibility of having a much more complex system based on rejection filters, which could be generalized to other uses (for example, rejection based on the energy or angle). We ended up steering away from that idea as it seemed a bit over-engineered, and our guess was that 99% of practical use-cases would just be simple spatial rejection, which is handled by a simpler design here.

[pshriwise]

@paulromano what would you think of having attributes for each domain type? This would allow rejection on a mixture of domain types if a user wanted to reject source sites for an entire universe and a few specific cells in another universe, for example.

[paulromano]

@pshriwise I did consider that, but it didn't seem to me that it would be worth the extra complexity. The case of combining a universe and some other cells could also be represented by using only cells.

[pshriwise]

@pshriwise I did consider that, but it didn't seem to me that it would be worth the extra complexity. The case of combining a universe and some other cells could also be represented by using only cells.

Hrmmm yeah that's a fair point -- either of the material or unvierse domain types can be represented equivalently using a set of cells. It's also consistent with how we handle volume calculations I suppose. I'm on board 😃

[pshriwise]

This feature is sure to be a crowd-pleaser 🎉 I believe we've gotten requests for this feature a number of times. Some small comments from me here.

[pshriwise]

Thank you @paulromano!