Multi-tumor/versioned analysis support

[briandoconnor]

The ES index built by merge_generated_metadata.py assumes 1) there's a normal and tumor and 2) there's only one run of each analysis. This is wrong, there can be multiple tumors per donor. Also, we need code that will handle multiple runs of analysis. If I have a newer analysis result (same workflow, newer datetime or version) then it should replace the previous analysis event in the index. Right now this is random which is used.

[briandoconnor]

Just to comment on this, this does not mean having multiple versions of the same analysis in the ES index, at least for now. Multiple versions of the same analysis will make the index very complicated. 9 times out of 10 we want the most recent version so we'll stick to that approach for now.