Add MAXO annotation evaluation

[caufieldjh]

No description provided.

[caufieldjh]

Not quite there yet - need to account for partial matches since these all test relations are stored as one-to-one triples but extracted are still one-to-many.
Also labeling isn't working as expected.
Also predicates may not be filtered correctly (all test triples have TREATS as predicate)

[caufieldjh]

This still has some strangeness going on with parsing of ungrounded results but otherwise works.
Needs more of the test cases to be included before it's ready.

[caufieldjh]

One remaining bit of weirdness: the extracted predicate is sometimes IS USED TO TREAT rather than TREATS, which is semantically identical but may be the tip of an iceberg of ungrounded predicate variations.

[caufieldjh]

Trying this out on all available test sets before merging. It will likely need some more work after that, but some of that work will concern improving the extraction template.

[caufieldjh]

One last bit of weirdness: the parsed triples can sometimes get populated with invalid values, despite starting as fully valid objects.
For example, this action_to_symptom relation:

        - subject: MAXO:0001175
          predicate: IS USED TO TREAT
          object:
            - end-stage liver disease
          qualifier:
          subject_qualifier:
          object_qualifier:

becomes this mostly-empty triple:

        - subject: end-stage liver disease
          predicate:
          object:
          qualifier:
          subject_qualifier:
          object_qualifier:

It shouldn't even be in the triples in the first place as the symptom isn't grounded to HP, so perhaps a boolean is flipped somewhere.