Distribute indexes of indexed access types first

[weswigham]

Fixes #27224

This change is a small addition (plus a bunch of comments to illustrate the process) to getSimplifiedIndexAccessType. To begin, let me explain the original issue. The contextual typing issue reported boils down to this:

function g<
    K extends "a" | "b">(x: ({ a: string } & { b: string })[K], y: string) {
    x = y;
}

This assignment worked in 3.0, and it should continue to do so. However the distribution change broke that, and in master (and 3.1) this assignment fails. Why? This is why: We'd see ({ a: string } & { b: string })[K] and distribute K into the intersection, producing { a: string }[K] & { b: string }[K]. Then for each we would get its apparent type, which means its constraint. That produces { a: string }["a" | "b"] & { b: string }["a" | "b"], which in turn becomes ({ a: string }["a"] | { a: string }["b"]) & ({ b: string }["a"] | { b: string }["b"]), which again becomes (string | unknown) & (unknown | string) which is simply unknown (or for constraint calculation, no constraint - which is subtly different in that no constraint means it's not assignable).

This changes this in two ways:

1. We distribute index types over the object type before distributing into the object type.
2. We do not perform distributions over the inner object type until the index type is no longer instantiable (and therefore might becomes a union we would need to distribute).

This changes how the algebra of the above example proceeds. Starting with ({ a: string } & { b: string })[K] we do not simplify via distribution, then we get its constraint, ({ a: string } & { b: string })["a" | "b"]. This distributes into ({ a: string } & { b: string })["a"] | ({ a: string } & { b: string })["b"] which becomes ({ a: string }["a"] & { b: string }["a"]) | ({ a: string }["b"] & { b: string }["b"]) which becomes (string & unknown) | (unknown & string) which is now just string (as unknown is a noop in an intersection like never is in a union), which is our desired type.

[weswigham]

@typescript-bot test this

[typescript-bot]

Heya @weswigham, I've started to run the extended test suite on this PR at 7cfd206. You can monitor the build here. It should now contribute to this PR's status checks.

[weswigham]

RWC build failed because master had some unaccepted diffs (which I have now accepted since they were just better error locations). The diff between master and this PR was null, so RWC is unaffected by this change.