Reevaluate Miri engine alignment checks

[RalfJung]

Two PRs will likely land soon that change the way the Miri engine does alignment checks: #63079 and #63075.

The former entirely disables alignment checks for CTFE (when the Miri engine is used as part of compilation). The latter moves the checking of "access validity" of a pointer (non-dangling and aligned) to the moment the pointer is dereferenced (using the * operator), as opposed to doing the checks only when a load/store actually happens.

As a consequence, the align field of MemPlace is now effectively dead weight in both instances of the Miri engine. However, Miri might need it again in the future if we decide to relax the rules for dereferencing pointers, which is a discussion I want to have after rust-lang/rfcs#2582. Then we should either remove the align field for good, or try to abstract it through the Machine trait so that CTFE does not have to carry it. Also see #63079 (comment).

Cc @oli-obk

[RalfJung]

@digama0 points out that an alternative future might be for the align field to move to PlaceTy, treating it conceptually more like part of the type of this place than its dynamic value. This would be good enough for what alignment is currently used for (justifying the alignment annotations on loads/stores); the sublte key difference lies in whether the dynamic alignment of trailing fields of unsized structs is factored in or not. (Currently it is.)

[RalfJung]

@digama0 points out that an alternative future might be for the align field to move to PlaceTy, treating it conceptually more like part of the type of this place than its dynamic value.

FWIW, that is exactly how I am treating place alignment in MiniRust.