bug in check_netcdf logic of PIOc_put_vars_tc

[jedwards4b]

The call to check_netcdf here
https://github.com/NCAR/ParallelIO/blob/master/src/clib/pio_getput_int.c#L1268
is in a section of code only used by iotasks. If the error handling is set to PIO_BCAST_ERROR
there is a call to MPI_Bcast https://github.com/NCAR/ParallelIO/blob/master/src/clib/pioc_support.c#L687
which cannot be completed because the comp_comm tasks are not present.

This results in a garbled error message at best. In the case I am working on I get:

MPI_BCAST, Message truncated on receive: An application bug caused the sender to send too much data

I suspect this is due to the comp_comm tasks arriving at a different MPI_BCAST call and trying to satisfy the collective.