Fix Corsair restart bug

[sandroos]

Corsair restarts were failing because multi IO units in VLSV were returning incorrect byte size for particles.

[rjarvinen]

We (@dphillips95) think there might be an issue with the Writer::multiwriteFlush function

vlsv/vlsv_writer.cpp

Line 505 in 0d06db7

bool Writer::multiwriteFlush(const size_t& counter,const MPI_Offset& unitOffset,

...in a special case when there's nothing to write from a process and outputType size is zero (also writeSize = 0) here:

vlsv/vlsv_writer.cpp

Line 552 in 0d06db7

MPI_Type_size(outputType, &writeSize);

...then MPI_File_write_at_all call hangs here:

vlsv/vlsv_writer.cpp

Line 569 in 0d06db7

if (MPI_File_write_at_all(fileptr,offset+unitOffset,multiwriteOffsetPointer,1,outputType,MPI_STATUS_IGNORE) != MPI_SUCCESS) {

One way to avoid this is to add an additional check if (N_multiwriteUnits > 0 && writeSize > 0) here:

vlsv/vlsv_writer.cpp

Line 565 in 0d06db7

if (N_multiwriteUnits > 0) {

and freeing outputType from MPI in the else block too when needed:

} else {
// Process has no data to write but needs to participate in the collective call to prevent deadlock:
if(N_multiwriteUnits > 0) {
     MPI_Type_free(&outputType); 
}

vlsv/vlsv_writer.cpp

Line 575 in 0d06db7

} else {

There could be a similar hang issue with addToBuffer here but haven't been able to test it:

vlsv/vlsv_writer.cpp

Line 588 in 0d06db7

addToBuffer(multiwriteOffsetPointer, writeSize, offset+unitOffset,outputType,comm);

[dphillips95]

Testing indicates that the above hanging issue with MPI_File_write_at_all when the datatype used has size 0 only occurs with Open MPI v4.1.x (or older?), v5.0.6 and 5.0.10 do not appear to have this issue any longer