Rollup merge of #160969 - Albab-Hasan:span-for-size-of-val-layout-errors, r=oli-obk

give layout errors from `size_of_val` and `align_of_val` a span

`size_and_align_of_dst` used the spanless `layout_of` so a `SizeOverflow` from `size_of_val` or `align_of_val` came out as

```
error: values of the type `[u8; 9223372036854775807]` are too big for the target architecture
```

and now points at the call:

```
error: values of the type `[u8; 9223372036854775807]` are too big for the target architecture
 --> src/lib.rs:5:53
  |
5 | pub fn check(x: *const Example) -> usize { unsafe { std::intrinsics::size_of_val(x) } }
  |                                                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
```

with `-Cdebuginfo=2` the error is hit earlier while describing the signature of the function so `get_function_signature` and `build_pointer_or_reference_di_node` forward the span `dbg_scope_fn` already computes same as https://github.com/rust-lang/rust/pull/145967 and https://github.com/rust-lang/rust/pull/146025 did for arrays and enums. the two revisions of the new test cover both.

this doesnt close https://github.com/rust-lang/rust/issues/25116. the vtable case in `tests/ui/limits/vtable.stderr` is emitted from a query keyed only on the type and the other builders in `debuginfo/metadata.rs` still drop spans.

cc @estebank who pointed at these two gaps on the issue.

r? @oli-obk