Fix multiple tree nodes with same name

[willdumm]

This PR is intended to fix an issue where multiple internal sequences may have the same node name.

It also adds a test in CollapsedTree._clade_tree_to_ctree which verifies that for any collapsed tree extracted from the history DAG, the maps from nodes to sequences and sequences to names are bijections.

[willdumm]

It turns out that with all the competing logic about node abundances, collapsed tree validation, etc, this is not as straightforward as I thought. I should be able to fix this properly as soon as I get back.

[willdumm]

This PR abandons any attempt to keep track of unobserved sequence names in the history DAG (they're there, but ignored).

The reason this is difficult is that ambiguous sequences are disambiguated in the DAG, and each disambiguated node that's created is given the same name as the original. This means that there are lots of different sequences associated with the same unobserved node name.

You'd think we could just rename all dag nodes after disambiguation, but this turns out to be tricky because the dag has some uncollapsed leaf-adjacent edges, and because of the logic which adds/removes a dummy leaf for unobserved, unifurcating root.

Because of all that, it's simpler to just fix the unobserved node names when collapsed trees are created. Note that this means unobserved node names will change when initializing a CollapsedTree.

Here are the guarantees:

• All sequences in a CollapsedTree will be uniquely associated to node names. That is, there can be no two sequences with the same name.
• Names are only unique in the scope of a single tree. In a CollapsedForest, if two unobserved internal nodes in different trees have the same name, they do not in general have the same sequence. However, observed sequence and naive sequence node names are preserved across all trees in the forest.
• If a CollapsedTree is extracted from a history DAG, and the maps from unobserved nodes to unobserved sequences to unobserved node names aren't bijections, then an error will be raised. For example, CollapsedTrees which contain separate nodes with the same sequence will cause gctree to crash (with an explanation). That's probably good, because such a tree cannot be maximally parsimonious. Note that this does not apply to the results of simulations, because simulated trees aren't stored in a history DAG.

[psathyrella]

This is awesome, thank you!