Handle memory better

[sandersky]

In some CircleCI builds the build VM has a lot of CPU's but running a worker on every CPU can cause the VM to run out of memory, due to Babel doing a lot of processing (one Babel instance per worker process). See if we can evaluate heap usage and not process files on every worker when heap usage is high. This way we can spawn all of the workers but can auto-adjust our processing in real-time to avoid running out of memory.