VPP should have option for writting in parallel

[aeslaughter]

Reason

Results from stochastic simulations are stored as VPPs. For large stochastic simulations
with billions of sub-applications it will be necessary to output the results in parallel (i.e., Nemesis for VPP files).

Design

Within the stochastic tools module the StochasticResults object, which is a VPP (see #14410), include operational modes: "replicated" and "distributed". I propose that this concept be added to the framework. The VPP object will have a "parallel_type" parameter that can be:

"manual" (not sure of the name)
This is the current behavior for VPP data and will be the default mode.

"replicated"
This mode assumes that the VPP data is computed in parallel but the data should be
replicated on each processor. The desired vector is assumed to be the parts on each rank gathered together in a single vector. Thus, when this mode is given this gather action is automatically performed by bringing the complete vector to the root process. The "_is_broadcast" flag will still be applied to make it truly replicated, if desired.

"distributed"
This mode assumes that the VPP data is computed in parallel but the data should remain
distributed. No, parallel communication if performed and the broadcast is disabled. However,
CSV output object (perhaps it is time for separate PP and VPP Output objects) will write N-rank
number of files: "input_out_vpp_name_0.csv.RANK" (???)

Impact

This is new capability and will enable the creation of large sets of VPP data, if needed.