Add documentation to facilitate estimating run time

[dmarx]

For each model:

• Size in RAM
• some model-standardized measure of inference speed

could make this stuff attributes on objects in https://github.com/dmarx/multi-modal-comparators