Experiment: can we improve key types we get from map iteration?

Currently, the key types are always something nested, confusing, and not very ergonomic.

## Custom struct instead of tuple
Maybe it would make sense to replace the tuple with some sort of a custom struct. That would let us be a little more descriptive with what's what, and provide helper methods.

## Avoid the wrapper in some cases
Maybe we can avoid some nesting when dealing with a terminal inner container, and so on. Research?

Start by tracing the map-iteration key types and how tuple nesting represents terminal inner containers. Compare a custom struct and reduced wrappers, then define the ergonomic target and verify that the chosen representation remains correct across the affected storage cases.

Written by the indexing model from the issue text.