Test driver fixes for error handling

[tagomoris]

This change follows #1302.
After #1302, I found that any errors in #write of output plugins are not raised in test case code like below:

assert_raise AnyErrors do
  d.run
    d.feed(events)
  end # events flushed and #write called - "AnyErrors" raised
end

Errors are shown on console, but we can't test it.
This change is to check all threads via thread plugin helper, and re-raise errors from these threads at the end of #run method. It shows all errors in all threads related with plugins explicitly as test results - so there's no need to show these errors on console.

Some changes are also included:

• add an option to retry for flush errors even in plugin development: to emulate write operations 2 or more times over failures
• use Process.clock_gettime instead of Time to work well even when timecop is used
• raise TestTimedOut error for hard timeout to show problematic test cases clearly
• change to wait post conditions after #run_actual, because output plugin test driver flushes and wait for completion of flushing in #run_actual

[tagomoris]

@repeatedly could you review this change?
This is required for both of migrations of exec plugins and forward plugin.

[tagomoris]

@repeatedly ping?

[repeatedly]

What happen with 0 limit?

[tagomoris]

It doesn't wait for running threads with 0.

[repeatedly]

LGTM