Dispatch manager for evaluating blocks of constitutive models

[reverendbedford]

We should have a manager routine that:

1. Maintains a discrete collection of devices (gpus, cpus, anything we can .to to).
2. Associates those devices with an "optimal" block size. We can eventually provide tools to find these optimal sizes, but for now they can be provided by the user.
3. Sets up a queue where the input is a big chunk of constitutive updates.
4. The manager routine assigns chunks to the devices until all the work is done and coordinates returning the assembled results to the original device.