Fix runner failures on travis

[pzrq]

IIRC Travis (and local) sometimes times out when downloading MongoDB. If this removes that then we should merge it and apply the pattern more generally so we don't have the sometimes fails, especially when a new version of MongoDB is released.

[pzrq]

Common example:

[pzrq]

Overall LGTM, thanks @durran 👍

This was cherry-picked from #49

For info, the commit originally passed on TravisCI and macOS Sierra, but failed in my Ubuntu 14.04 VM.

I resolved the VirtualBox Ubuntu VM issue by some combination of rm -rf node_modules; npm install, find ~ -name mongodb-linux-x86_64-3.2.11.tgz and deleting those, and killall mongod. Restarting the Ubuntu VM did not work.

  27 passing (729ms)
  2 failing

  1) fetch() local "before all" hook:
     Uncaught MongoError: failed to connect to server [localhost:27017] on first connect
      at Pool.<anonymous> (node_modules/mongodb-core/lib/topologies/server.js:313:35)
      at Connection.<anonymous> (node_modules/mongodb-core/lib/connection/pool.js:271:12)
      at Socket.<anonymous> (node_modules/mongodb-core/lib/connection/connection.js:165:49)
      at emitErrorNT (net.js:1276:8)
      at _combinedTickCallback (internal/process/next_tick.js:74:11)
      at process._tickCallback (internal/process/next_tick.js:98:9)

  2) fetch() local "after all" hook:
     TypeError: Cannot read property 'drop' of undefined
      at Context.<anonymous> (test/fetch.test.js:32:17)

npm ERR! Test failed.  See above for more details.

The VirtualBox Ubuntu VM allowed me to drop the Processor execution cap to 10%, and confirm that after this fix other tests will warn about speed but not fail, and further that going back to master before this fix does trigger the 10s timeout error.