Why Does the Syntax (AST) Depend on the Parser Instead of the Other Way Around?

[ds1sqe]

I've been diving into the architecture of rust-analyzer but syntax crate seems to have a dependency on the parser crate like upper picture

I would expect the parser to rely on the AST to build a representation of the source code, not the other way around.

Could someone explain why this design choice is made? Is there a specific reason or advantage.
Any insights or examples would be greatly appreciated!

Thank you!