Ensure error recovery does not misparse in the event of ambiguous syntax

[OmarTawfik]

even though a NamedArgument is required inside {}, it is matched successfully as a post-fix expression, and thus, failing for expecting the try body afterwards.

Originally posted by @OmarTawfik in #853 (comment)

[Xanewok]

@OmarTawfik this refers to the Separated parser of NamedArguments at (latest) https://github.com/NomicFoundation/slang/pull/853/files#diff-af454bb70b3f914baf6e383790278f8efd4f555a445ec8b915164e93d73cc1d1R3354, right?

[Xanewok]

Here's more context about the underlying issue:

It includes a delimited parser ({ NamedArguments }) that recovers on the empty content at the nearest delimiter matching the current tree, so the parse returns "Match, but recovered" (includes SKIPPED nodes).

Now, since the expression parser will happily tack that recovered group onto the current accumulator, we end up with a.b() ++ {} (recovered).

Because this syntax is generally ambiguous and is explicitly disambiguated by solc with a token lookahead to check for at least a single item (Identifier followed by a colon), we can't really attempt recovering here from an empty match as this leads to the issue where we can recover past the point where another parser (Block here) should pick up, leading to a misparse.

We already have a notion of RecoverFromNoMatch for recovery that's currently always emitted for terminators; we should surface this to a DSL and allow the users to disable error recovery in the face of ambiguous syntax such as this.

[OmarTawfik]

@Xanewok fixed the title. Thanks!

[Xanewok]

Changed the title to better reflect the underlying issue and what we're trying to fix in the short/mid term here.