[reports] Speed up report generation significantly

[jonathanmetzman]

Reduce report generation time from minutes to ~20 seconds. These measures which rely on multiprocessing won't be used on the dispatcher to avoid overloading them. But they should help local report generation a lot.

1. For differential coverage, use multiprocessing if requested to download coverage data from GCS.
2. For report plots, use multiprocessing to generate plots and cache them in parallel so that rendering with jinja is instantaneous.

[inferno-chromium]

@jonathanmetzman - this would be nice to get landed, fast report generation is awesome. i am fine with some repetitions.