Improve mobile gaming example user experience

Few things I noticed while running mobile gaming example that could be improved:

1. When running on direct runner the default input is too large (20G), so it seems as though the pipeline is stuck without an progress updates or metrics.
This could be improved by using a much smaller dataset by default.

2. Even when running on dataflow runner, with default worker settings and auto scaling, it still takes more than 30 minutes to run. We could use a much smaller dataset here too.

Also the documentation of these examples could be improved in both the code docstring as well as beam quick start guide.

Imported from Jira [BEAM-2286](https://issues.apache.org/jira/browse/BEAM-2286). Original Jira may contain additional context.
Reported by: vikasrk.

Start with the mobile gaming example and the Beam quick start guide, then compare its default input and runtime on the direct runner and Dataflow runner. Done means the defaults use a smaller dataset that provides timely progress or metrics, completes faster on Dataflow, and the example docstring and quick start guidance reflect the updated behavior.

Written by the indexing model from the issue text.