cross eval methodology

[sethtroisi]

Per offline discussion around https://github.com/sethtroisi/minigo/pull/4 and https://cloudygo.com/all-eval-graphs?bucket=cross-eval

Andrew suggested

• Removing all models below the "knee" (e.g. model 200)

Seth mentioned

• [ ] Completely removing ratings.py and all that rsync jazz
• [ ] Add a models cloud bigtable table with models and runs (#797)
• [ ] clean up cross-eval panel/model selection.
• [ ] Figure out how to correctly tag run (e.g. v15) in cbt

[sethtroisi]

• [x] Fix length mismatch between cc eval and backfill (they are both strings instead of ints now)