Explicitly encode keyof behaviors for never and unknown into getIndexType

[weswigham]

keyof never -> string | number | symbol (changed from never)
keyof unknown -> never (unchanged, but no longer derived from unknown's empty apparent type)

The reasoning for the first is pretty simple - we know keyof is contravariant, so the more specialized the input type is, the broader the output type is. For each key we add on the input type, we specify a smaller and smaller set of inputs, while the output set of keys grows. Once the input is the empty set of types (never), our result should encompass every possible key.

Fixes #30657

[weswigham]

@typescript-bot test this & @typescript-bot run dt - I honestly have no idea what this will affect in the wild. Historically people haven't really relied on incorrect boundary relations in TS, so we shall see.

[typescript-bot]

Heya @weswigham, I've started to run the extended test suite on this PR at d147a0d. You can monitor the build here. It should now contribute to this PR's status checks.

[typescript-bot]

Heya @weswigham, I've started to run the Definitely Typed test suite on this PR at d147a0d. You can monitor the build here. It should now contribute to this PR's status checks.

[jack-williams]

This is great change IMO. One semi-related question. Is there a reason for the following, mainly regarding number:

type A = keyof Object; //  "constructor" | "toString" | "toLocaleString" | "valueOf" | "hasOwnProperty" | "isPrototypeOf" | "propertyIsEnumerable"
type B = keyof number; //  "toString" | "toFixed" | "toExponential" | "toPrecision" | "valueOf" | "toLocaleString"
type C = keyof object; //  never;

[weswigham]

@RyanCavanaugh @ahejlsberg no DT or RWC changes 😸