Inaccurate memory usage estimation

[calebwin]

There are several aspects of this:

• ScaleBy constraint on a task level, PT level (for Grouped/Blocked), and for each split/merge/cast function
• Owned constraint (for mutation) and also ensuring that sample-related values are updated properly in case of mutation
• Constraints that can be applied to all PAs for a task but then fused together only for the same PA for ScaleBy
• Properly estimating memory usage in optimize_fit_in_cache for fused tasks where some values don't need to be in cache for the whole stage