CI / Travis running close to Xmx limit during Scaladoc

[lrytz]

Looking at more travis logs lately, I saw quite a bunch of messages like this one during Scaladoc generation:

[warn] In the last 9 seconds, 5.829 (65.4%) were spent in GC. [Heap: 0.73GB free of 1.83GB, max 1.83GB] Consider increasing the JVM heap using `-Xmx` or try a different collector, e.g. `-XX:+UseG1GC`, for better performance.

https://app.travis-ci.com/github/scala/scala/jobs/635630711