[hari2353]

Adds bound_plan::write::is_write_statement using only the clause-list parser (no legacy AST fallback), and switches the embedded dispatcher to it.

[genezhang]

Thanks for looking at #1447. I compared this against main and could not find a statement where the two behave differently, so I'd suggest not merging it as is.

What I did. I compared StatementTargets::of(text).is_write(), which is what the dispatcher reads today, with the new is_write_statement(text) on 25 statement shapes: plain reads and writes, a write in a later UNION arm, COUNT { } in SET / RETURN / WHERE of a write, MERGE, FOREACH, CALL, DETACH DELETE, SET n:Label, UNWIND … CREATE, and more. The results are identical for all 25.

Why. StatementTargets::of already tries the clause-list parser first and returns its result when the text parses (#1504). The legacy parser is only the fallback for text the clause-list parser rejects, and none of my shapes reached it. The UNION half of #1447 was fixed by #1504. For the COUNT { } half, the clause-list parser now reads those statements, so they already classify as writes on main. I checked the classification only, not a full COUNT { } write executed on chdb.

So the change is a refactor, and it duplicates the Create | Set | Remove | Delete match that prepared.rs already has. If the goal is a single place for that decision, one option is to have StatementTargets::of call is_write_statement. Either way it would need a test (for example the COUNT { } write shapes through the dispatcher).

I'd say #1447 can be closed once that COUNT { } write is confirmed end to end on chdb. If you'd like to take that on instead of this refactor, we'd welcome it.