Direct Runner should marshal data in a similar way to Dataflow runner

I would test my pipeline using the direct runner, and it would happily run on a sample. When I ran it on the Dataflow runner, it'll run for a hour, then get to a stage that would crash like so:
 
> java.util.concurrent.ExecutionException: java.lang.RuntimeException: Error received from SDK harness for instruction -224: execute failed: panic: reflect: Call using main.HistogramResult as type struct \{ Key string "json:\"key\""; Files []string "json:\"files\""; Histogram palette.ColorHistogram "json:\"histogram,omitempty\""; Palette []struct { R uint8; G uint8; B uint8; A uint8 } "json:\"palette\"" } goroutine 70 [running]:>
> This was because I forgot to register my HistogramResult type.
> It would be useful if the direct runner tried to marshal and unmarshal all types, to help expose issues like this earlier.
> Also, when running on Dataflow, the value of flags, and captured variables, would be the empty/default value. It would be good if direct also caused this behaviour. For example:
> ```

prefix := “X”
s = s.Scope(“Prefix ” + prefix)
c = beam.ParDo(s, func(value string) string {
return
prefix + value
}, c)

```

Will work prefix "X" on the Direct runner, but will prefix "" on Dataflow. Subtle behaviour, but I suspect the direct runner could expose this if it marshalled and unmarshalled the func like the dataflow runner.

Imported from Jira [BEAM-6372](https://issues.apache.org/jira/browse/BEAM-6372). Original Jira may contain additional context.
Reported by: bramp.