Document best-practise for running yapf (in particular: for performance)

The current documentation gives no opinion on how to efficiently run yapf.
Large, recursive re-runs are very slow.
Some opinionated documentation of how to optimise running would be very useful (perhaps calling a distinction between runs on CI and runs on developer machines).

This may be aided by profiling to determine how/where to optimise.
other ideas:
- tooling/config to detect git-changes and only re-yapf changed files
- a runner to split targets across CPUs