Calculate allocated memory based on requirements to reduce GPU load

[KTZ228]

To reduce the GPU RAM load, we can consider calculating the allocated RAM based on expected requirements instead of setting a default value that can be too high (as it is done now in single_subject_pipeline_with_slurm.m).

I'm unsure how to go about this, but if we get this to work we could greatly reduce the RAM load on the GPUs.
Making this improvement is only worth it, however, if the high load we put onto the GPUs is caused by our high RAM usage.