Change how we handle SemMedDB edges in ranking

[finnagin]

SemMedDB seems to be returning lots of odd edges and bad results that get pushed higher in rankings.

Lots of potential options to address this:

• [x] Use subject and object confidence scores in ranking
• [ ] Use subject and object novelty in ranking
• [x] Condense SemMedDB edges into one edge
• [ ] SemMedDB antonym handling

[finnagin]

Should look into averaging semeddb edge publication counts using:

• harmonic mean
• geometric mean
• median
• arithmetic mean
• L-infinity

[dkoslicki]

On branch issue1695. Should test out the different averaging methods when combining multiple SemMedDB edges and see which ones we like. Issue #1684 is needed for the other items

[dkoslicki]

Closing as @mfl15 's approach for filtering SemMedDB will likely fix this issue