Fix ConstantFieldPropagation signed packed field handling and improve Heap2Local's

[kripken]

CFP already had logic for truncating but not for sign-extending, which this
fixes.

Use the new helper function in Heap2Local as well. This changes the model
there from "truncate on set, sign-extend on get" to "truncate or sign-extend
on get". That is both simpler by reusing the same logic as CFP but also more
optimal: the idea to truncate on sets made sense since sets are rarer, but if
we must then sign-extend on gets then we can end up doing more work
overall (as the truncations on sets are not needed if all gets are signed).

Found by #6486

[tlively]

How do signed gets make that more complicated? If you truncate the value on the set and sign extend it again on the get, you'll still get the correct value back. Can't we leave the redundant truncation and sign extension to be cleaned up by other ops instead?

[kripken]

It's more complicated in that it's hard to see what is best here. That is, correctness is easy but we'd need to count how many gets are signed and how many are unsigned, to know which model is more efficient. So it is simplest to do the simple thing (fix up reads in one place), and leave optimizing that further to another pass.

(Atm nothing optimizes this further, but we could do it - basically count the sign-ext operations on local.gets etc. and maybe decide to shift from fixups on gets to sets or vice versa. We actually do something similar for linear memory loads and stores, but not locals.)

[kripken]

Or, to put it another way, this PR simplifies Heap2Local code while also making it more efficient in at least one case (the case where there are many signed reads and many writes: we did the worst possible thing in that case before, which was to do a fixup on both those reads and those writes).

[tlively]

Makes sense 👍