Add typedef declaration space, unify typedef name gathering

[weswigham]

And strengthen errorUnusedLocal to no longer error when a declaration name cannot be found.

Fixes #18097

There were actually two separate but related crashes in that issue - one because getDeclarationSpaces did not contain a clause for jsdoc typedefs, and one because getNameOfDeclaration did not know how to find the name for a jsdoc typedef when its own .name field was not provided - I've unified the codepath used by the binder to get the name with the path used in getNameOfDeclaration, so minimally if the symbol is bound, then it should be nameable now. In doing so I found that we had some test cases which were unnameable but bound or nameable but unbound, and added clauses to the new getNameOfJSDocTypedef based on where I saw either naming or binding occur prior to this change, so both should always be possible in both cases now. To do this, I wanted to ensure I was being exhaustive with the nodes checked; so I've also separated the jsdoc node properties from the Node interface and created a HasJSDoc union which represents all nodes which the parser applies JSDoc to. In doing so, I actually found a few types where we could have been applying jsdoc, but were not - which we now are.

[DanielRosenwasser]

You could make fail return never and write return fail(message, stackCrawlMark || assertNever)

[ghost]

It's nice to check that the type of the member is never though. Or did you just mean to change the implementation of assertNever instead of to replace it?

[weswigham]

I think you just mean that throw "Unreachable"; is not required if fail just returns never - which seems fine to me; so I've done that.

[mhegazy]

& Node ?

[weswigham]

All members of HasJSDoc are already nodes. The interface HasJSDocNodes would require that. Is that confusing? Should I rename the second to something like JSDocContainer?

[mhegazy]

JSDocContainer seems like a better name