Optimize sum-type storage

Right now, when we lower sum types they (roughly) get expanded into a flattened product of all fields appearing in all cases, with only the storage corresponding to the active case actually getting initialized and used at run-time.

But this approach is highly wasteful! In particular an ADT such as
```hs
data ColorfulFloat =
RedFloat Float
BlueFloat Float
GreenFloat Float
```
should only use 5 (1+4) bytes of storage, not 13 (1+4+4+4)!

To improve this situation, we need to modify the Imp codegen to reuse storage already allocated for other cases while emitting the destinations for sum types.

Start by reading the Imp codegen path that emits destinations for sum types, then trace how storage is allocated for each case. The work is complete when inactive case fields no longer add separate storage and an example such as ColorfulFloat uses shared storage for its cases.

Written by the indexing model from the issue text.