Implement do

[mscroggs]

Implement a do algorithms that takes a graph and replaces a node with a constant ParameterNode and removes any edges going into that node.

As expectation and other algorithms already only iterate over ancestors of output nodes, there's no need to trim the graph further when applying a do.

Builds on #39, so is currently set to merge into that branch.

[willGraham01]

Pre-emptive note here: in #40 the ParameterNodes are always assumed to have values that will vary (see how parameter nodes are gathered here).

To avoid the solver accidentally picking up these constant parameters as values that should vary we could either:

• ParameterNodes have another internal is_constant flag, that defaults to False but can be set to True. Then the aforementioned line checks if node.is_parameter and not node.is_constant.
• ConstantNode class like was suggested before (but I think that's overkill since it would be almost identical to ParameterNode).
• Have the do operator edit the DistributionNode's that now depend on the constant node, by moving the now-fixed value from parameters to constant_parameters for each affected node.

[mscroggs]

• Have the do operator edit the DistributionNode's that now depend on the constant node, by moving the now-fixed value from parameters to constant_parameters for each affected node.

Ooh, we could just remove the node that the do is setting the value of and do this move without needing to add a new node. This feels very neat so I'll do this

[willGraham01]

I think the new_nodes dictionary in the do method is being ignored? Or it doesn't need to exist perhaps? Either way I think I get what it's meant to be doing.

With that fix - and some slightly more verbose language - I think do should be do-ing what we expect 🥳 Have added some suggestions for what to test regarding this method - TLDR would focus on only testing the connectivity of the graph that do produces, and that the parameters have been moved to constant_parameters correctly.

[mscroggs]

I think the CI will pass here once #56 is merged