revert 74989 ("Implement Index and IndexMut for arrays")

[matthiaskrgr]

This PR reverts #74989 and adds a testcase for

const fn foo() {
    let mut array = [[0; 1]; 1];
    array[0][0] = 1;
}

which compiles on stable and beta but not on nightly due to #74989 .
The offending commit had a crater run but the breaking change was missed which means it probably would not have landed if the breaking change had been noticed in time.

This PR fixes compilation of real-world crates such as https://github.com/alacritty/vte/

Fixes #79152

[rust-highfive]

r? @Mark-Simulacrum

(rust-highfive has picked a reviewer for you, use r? to override)

[Aaron1011]

This will break any code that started relying on the (insta-stable) Index and IndexMut impls.
I think the targeted fix suggested by @lcnr would be a better idea: #79152 (comment)

[Aaron1011]

I've opened #79427 - the issue turned out to be a missing call to resolve_vars_if_possible

[oli-obk]

we decided in last weeks compiler team meeting that we'd do the revert and then immediately open a PR with the fix and impl again. I guess that only makes sense if we crater it, considering that there's a fix PR open. Since we didn't discuss running crater, and we do have a fix PR, let's favour the fix PR.

cc @rust-lang/compiler I'm making a decision against the meeting decision, since I think there's little benefit in reverting, only to readd with the fix and merge that immediately after.