Update lib dom for 4.2

[orta]

Updates the dom.d.ts files - Key changes:

• Some of the big DOM querying APIs default to HTMLElements (shoudn't be a breaking change)
• Some of the DOM query APIs are now generic, so they can work with SVG or XHTML instead of just HTML DOMs
• import.meta.url is now defined

Breaking changes

• DOM APIs which are deprecated, are now marked as deprecated
• The Streams API had some fields removed which aren't used in any browsers

microsoft/TypeScript-DOM-lib-generator#960

[orta]

OK, merging this at the end fo the day and I'll be keeping an eye out over the week for changes

[ahejlsberg]

This PR breaks a lot of DT tests. In particular, lots of noise about modifiers of some url property not matching. Also, there's this new definition of getElementById:

getElementById<E extends Element = HTMLElement>(elementId: string): E | null;

Type parameters that aren't used in the parameter list are an anti-pattern. Using a type parameter only in the return type basically amounts to an implicit type cast to the contextual type, if any. In this case the change is also breaking because the new definition produces Element when inference from the contextual type infers a type that isn't assignable to Element, such as:

declare function getElementById<E extends Element>(s: string): E | null;
declare function foo(x: string | HTMLElement): void;
foo(getElementById('hello'));  // Now errors

What happens here is that we infer string | HTMLElement for E and since that isn't assignable to Element, we revert to the constraint Element, which then fails.

[ahejlsberg]

BTW, we typically run the DT tests before merging PRs like this. Having lots of breaks in the DT tests in master leads to a bunch of counter productive noise in other PRs.

[orta]

Oof, yeah, my bad - sorry, my first time sending them over from the DOM repo.

I can revert tomorrow and look at extracting the Element changes out to get a safer merge