Improve InstrumentationTestRunner exception handling.

This commit fixes two somewhat related problems:
  - Attempting to run a method which does not exist caused a runtime
exception which stopped the test run. Change this so the runner reports
an individual test failure instead
  - A runtime exception during the test run would cause it to stop
completely, with no information dumped to the logcat or stdout. Now
exceptions are trapped and reported to stdout.

Also added associated unit tests to test these two conditions.

Related bug 2812262.

Change-Id: I383f9b9bad99f14cb51071800fa9bdbf6a6a1119