AsyncResult and px: groupby='first', groupby='merge'

The current groupby argument always displays the result of all engines.

It is useful sometimes to only display the result of a single engine -- when the execution / result is known to be symmetric. I propose adding a groupby argument:

- `first` : only display the result of the engine id 0; results from other engines are consumed and abandoned.

- `merge` : only display the unique results. For example, if engine 0 to 3 have identical results, and engine 4-8 have another identical results, then a two sets of results are displayed. The interactive debugger from intel mpi (a wrapper of gdb with result-merging) has this feature.

The merge mode is much harder in this case, because 'identical' is difficult to define -- objects have been serialized and deserialized; id is different. Using hash may be a possibility.

The `first` mode will already be very handy in trimming down the verbosity in a lot of cases.