Demonstrate splitting struct into individual scalars

[VictorYing]

This us a fairly mechanical transformation to the generated C++:

• Add local variable definitions at the top of the lambda body to extract the fields from the struct.
• Avoid any direct use of the struct elsewhere in the lambda body: replace any access to the fields of the struct to an access to the corresponding local scalar variable.

After this transformation, SCC/T4 does a slightly better job packing registers and deciding how to transfer values between the fine-grained tasks created by SCC/T4 within the lambda body, and this manages to avoid one instance of a mem_runner (i.e., transfer via the heap) for one task spawn associated with the for loop inside of case 1.