fix deadlock in asyncSend

[psychocoderHPC]

• GridBuffer.hpp and ParticlesBuffer.hpp: set gpuFree event to communication event
• close Freeze of picongpu if used MPI_Allgather in plugin #1322 Freeze of picongpu if used MPI_Allgather in plugin

A deadlock in the async communication is triggered when process one opened a MPI_Receive while the destination process two not opened the MPI_Send and starts a MPI collective operation. In this case the event system stops and not start the MPI_Send from process two which wait in a side dependency chain. Process one wait for finish of the MPI_Receive before it can start the collective operation. Now we had a classical deadlock.

This patch only solve the deadlock a cleanup of the interfaces will done in a separate pull request.

Tests:

• example plugin from Freeze of picongpu if used MPI_Allgather in plugin #1322 ( command: mpiexec -n 4 ./picongpu -d 2 2 1 -g 64 128 32 -s 10 --mpitest.period 1 --mpitest.mode 3)

[ax3l]

thank you a lot! ✨