Don't use babel when running tests using karma

[foolip]

Currently instrumentation is running after babel, which results in
incorrect and nonsensical coverage information. The node vs. web
coverage here is an example:
https://codecov.io/gh/GoogleChromeLabs/confluence/src/f6802df5a6be89042a1db1e7b82f0a0a90906caf/lib/dao/indexed_dao.es6.js

This could be fixed by changing the order:
#404

However, it's not clear if that's correct:
webpack-contrib/istanbul-instrumenter-loader#86

There's no need to run Babel to test stable versions of Chrome or Firefox,
so just skip babel-loader.

[codecov-io]

Codecov Report

Merging #398 into master will decrease coverage by 7.06%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #398      +/-   ##
==========================================
- Coverage   60.77%   53.71%   -7.07%     
==========================================
  Files          54       54              
  Lines        2904     1670    -1234     
==========================================
- Hits         1765      897     -868     
+ Misses       1139      773     -366

Flag	Coverage Δ
#node	73.74% <ø> (ø)	⬆️
#web	55.16% <ø> (-0.4%)	⬇️

Impacted Files	Coverage Δ
lib/property.es6.js	50% <0%> (-16.67%)	⬇️
lib/u2/DAOController.es6.js	6.25% <0%> (-12.8%)	⬇️
lib/dao/indexed_dao.es6.js	60% <0%> (-8.43%)	⬇️
lib/web_apis/release.es6.js	41.66% <0%> (-8.34%)	⬇️
lib/object.es6.js	9.09% <0%> (-7.58%)	⬇️
lib/dao/local_json_dao.es6.js	12.5% <0%> (-7.02%)	⬇️
lib/u2/ScrollDAOTable.es6.js	4.34% <0%> (-6.18%)	⬇️
lib/client/state.es6.js	10.34% <0%> (-5.45%)	⬇️
lib/client/api_service.es6.js	29.72% <0%> (-5.28%)	⬇️
lib/confluence/metric_computer_runner.es6.js	75.38% <0%> (-5.27%)	⬇️
... and 37 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f6802df...dae38f1. Read the comment docs.

[foolip]

Self-review because this is a subset of #395 already approved by @Hexcles

[foolip]

webpack-contrib/istanbul-instrumenter-loader#86 is an issue about this that I think is related to the problem.

[foolip]

The change in coverage is of course because of the semi-random offsets of line coverage before, which would incorrectly increase line coverage.