Sebastian Graf authored Mar 31, 2023 and Marge Bot committed Aug 09, 2023

We eliminate EmptyCase by way of `coreToStg (Case e _ _ []) = coreToStg e` now.
The main reason is that it plays far better in conjunction with eta expansion
(as we aim to do for arguments in CorePrep, #23083), because we can discard
any arguments, `(case e of {}) eta == case e of {}`, whereas in `(e |> co) eta`
it's impossible to discard the argument.

We do also give the same treatment to unsafeCoerce proofs and treat them as
trivial iff their RHS is trivial.

It is also both much simpler to describe than the previous mechanism of emitting
an unsafe coercion and simpler to implement, removing quite a bit of commentary
and `CorePrepProv`.

In the ghc/alloc perf test `LargeRecord`, we introduce an additional Simplifier
iteration due to #17910. E.g., FloatOut produces a binding
```
lvl_s6uK [Occ=Once1] :: GHC.Types.Int
[LclId]
lvl_s6uK = GHC.Types.I# 2#

lvl_s6uL [Occ=Once1] :: GHC.Types.Any
[LclId]
lvl_s6uL
  = case Unsafe.Coerce.unsafeEqualityProof ... of
    { Unsafe.Coerce.UnsafeRefl v2_i6tr -> lvl_s6uK `cast` (... v2_i6tr ...)
    }
```
That occurs once and hence is pre-inlined unconditionally in the next Simplifier
pass. It's non-trivial to find a way around that, but not really harmful
otherwise. Hence we accept a 1.2% increase on some architectures.

Metric Increase:
    LargeRecord