Revert binding pattern inference changes, but only for tuples

[andrewbranch]

Partial revert of #45719—we now let an array binding pattern turn inference of an array into a tuple, which is really nice. I still think the behavior we were seeing for object binding patterns was undesirable, and #45846 supports that hypothesis since all the related errors were due to a tuple turning into an array.

[andrewbranch]

@typescript-bot user test this inline

[typescript-bot]

Heya @andrewbranch, I've started to run the inline community code test suite on this PR at c3bcc4c. You can monitor the build here.

Update: The results are in!

[typescript-bot]

@andrewbranch
Great news! no new errors were found between main..refs/pull/46009/merge

[andrewbranch]

We realized that

declare function f<T>(): T;
const [a, b, c] = f();

should really error, and this PR makes it legal (T is inferred as [any, any, any]). While it may be true that object binding patterns never contribute anything useful to type argument inference, it is not true that array binding patterns always contribute usefully to type argument inference. What we really want is for binding patterns to contribute contextual typing information to other inference sources, while never being allowed to stand on their own as inference sources. I’m going to close this and do a full revert of #45719 so we can investigate a more thorough solution in the future.