Remove more redundant predictions and curate some DOID mappings

[bgyori]

This PR adds manual curations for some of the DOID mappings remaining after #79.

I also realized that EFO also provides mappings to DOID and these aren't symmetric with what DOID provides so have to be independently considered. So this PR removes all the DOID-EFO mappings where EFO already provides a mapping to DOID for a given EFO ID. This leaves only 137 remaining DOID-EFO mappings that are potentially useful.

[cthoyt]

I'm not sure if I agree with removing the manually curated mappings from DOID to EFO even though they are redundant of some available from EFO to DOID. I think having evidence for the mapping being true both ways is nice, especially because DOID->EFO mappings and EFO->DOID mappings are curated by two different groups of people (with Biomappings being a potential third group of people). Mappings probably need an assembly/deduplication step between curation in Biomappings and actual usage

[bgyori]

We can definitely discuss this, my interpretation is that EFO and DOID consider cross references to each other exact matches and so it is still redundant to manually curate mappings and make them available independently here. But I might be missing something.

[bgyori]

What I meant above by "not symmetric" is just that they haven't incorporated each other's mappings but otherwise all the mappings seemingly are meant to be exact one-to-one.

[cthoyt]

I guess that both EFO and DOID use the same semantics when talking about cross references (though it might be using dbXref and not a more specific relation, so it could be hard to say). I'll try to do an assessment to see how consistent these mappings are going both ways