Process parse errors (including error nodes) in TextDocumentState

[sungshik]

Before this PR, parse trees were processed twice to find error nodes (i.e., the exact same computation was repeated a second time): first in TextDocumentState (to figure out if lastTreeWithoutErrors should be updated), and next in RascalTextDocumentService (to translate error nodes to LSP diagnostics).

After this PR, all processing of parse errors (including error nodes) is moved out of RascalTextDocumentService into TextDocumentState. There is a new API call getCurrentDiagnosticsAsync (optionally with a debounce delay) for consumers to get diagnostics, similar to the existing getCurrentTreeAsync.

[sungshik]

Additional comments

[PieterOlivier]

Looking good! Maybe you can move the Debouncer class to its own file?

[PieterOlivier]

As a <tree,diagnostics> pair seem to belong together (not suprisingly), would ik make sense to have one Versioned object with a version and a pair of <tree,diagnostics>?

[sungshik]

In principle, yes, it would. However, in many places in the project (including the Parametric... classes), a Versioned<ITree> is currently expected, so it seems quite an invasive change to supply a Versioned<Pair<ITree,List<Diagnostic>> instead. I'm not worried about the complicated type parameter btw (i.e., we could hide the structure with an auxiliary class). But it's just a lot of changes to places that we don't really need/hope to touch right now 😉.

At the same time, I imagine the reporting of diagnostics can be streamlined a bit more if the diagnostics become accessible in some places where currently there's only a parse tree. Tt requires a bit more thought, though. I think I'd prefer to do that in a separate PR.