If PetaVision was compiled without GPU acceleration, it is now an error to set receiveGpu (in connections) or updateGpu (in layers) to true. Previously, the values in the params file would be read and discarded; and in the files that got saved in checkpoints and the output path, those parameters would be set to false.

The rationale for this change is to help catch situations where the user meant to use GPU acceleration, but failed to turn PV_USE_CUDA on. By throwing an error immediately, we avoid running for a long time before realizing that the GPUs are not being used.