Improve terminal output when running frontend tests in "watch" mode.

The following discussion from gitlab-ce!29560 should be addressed:

• @pslaughter started a discussion: (+5 comments)

  Here's a demo of running yarn test app/assets/javascripts/ide --watch.

  20190613_yarn_test_watch_demo

  I think this is a good iteration, but are we okay with the following behavior:

  • It's possible (but unlikely because of how slow karma is) that the output of jest is intertwined with karma.
  • I lose my jest errors because the karma output is so large.

  ---

  I wonder if we could somehow capture error lines (and their context) from the output and append those lines to the bottom after both scripts have run 🤔...

  I'd like your thoughts @winh, @leipert, and @tristan.read too 😄