Previously we would rebuild all drops on every early exit from a scope, which for code like:

```rust
match x {
    a => return 1,
    b => return 2,
    ...
    z => return 27
}
```

would produce 27 exactly same chains of drops for each return, a O(n*m) explosion in drops.

Use it instead of a `panic` for inexhaustive matches and correct the
comment. I think we trust our match-generation algorithm enough to
generate these blocks, and not generating an `unreachable` means that
LLVM won't optimize `match void() {}` to an `unreachable`.

this introduces a DropAndReplace terminator as a fix to #30380. That terminator
is suppsoed to be translated by desugaring during drop elaboration, which is
not implemented in this commit, so this breaks `-Z orbit` temporarily.

Primarily affects the MIR construction, which indirectly improves LLVM
IR generation, but some LLVM IR changes have been made too.

* Handle "statement expressions" more intelligently. These are
  expressions that always evaluate to `()`. Previously a temporary would
  be generated as a destination to translate into, which is unnecessary.

  This affects assignment, augmented assignment, `return`, `break` and
  `continue`.
* Avoid inserting drops for non-drop types in more places. Scheduled
  drops were already skipped for types that we knew wouldn't need
  dropping at construction time. However manually-inserted drops like
  those for `x` in `x = y;` were still generated. `build_drop` now takes
  a type parameter like its `schedule_drop` counterpart and checks to
  see if the type needs dropping.
* Avoid generating an extra temporary for an assignment where the types
  involved don't need dropping. Previously an expression like
  `a = b + 1;` would result in a temporary for `b + 1`. This is so the
  RHS can be evaluated, then the LHS evaluated and dropped and have
  everything work correctly. However, this isn't necessary if the `LHS`
  doesn't need a drop, as we can just overwrite the existing value.
* Improves lvalue analysis to allow treating an `Rvalue::Use` as an
  operand in certain conditions. The reason for it never being an
  operand is so it can be zeroed/drop-filled, but this is only true for
  types that need dropping.

The first two changes result in significantly fewer MIR blocks being
generated, as previously almost every statement would end up generating
a new block due to the drop of the `()` temporary being generated.

while I'm at it, remove the "extra caching" that I was doing for no good
reason except laziness. Basically before I was caching at each scope in
the chain, but there's not really a reason to do that, since the cached
entry point at level N is always equal to the last cached exit point
from level N-1.

also, when exiting a scope, assign the final goto terminator with the
target scope's id

It's nice to be able to index with a scope-id,
but coherence rules prevent us from implementing
`Index<ScopeId>` for `Vec<ScopeAuxiliary>`, and I'd
prefer that `ScopeAuxiliary` remain in librustc_mir,
just for compilation time reasons.

This was triggered by me wanting to address a use of DUMMY_SP, but
actually I'm not sure what would be a better span -- I guess the span
for the function as a whole.

In MIR we previously tried to match `let x in { exprs; let y in { exprs; }}` with our data
structures which is rather unwieldy, espeicially because it requires some sort of recursion or
stack to process, while, a flat list of statements is enough – lets only relinquish their lifetime
at the end of the block (i.e. end of the list).

Also fixes #31853.

TODO: find a correct borrow region

Fixes #31482