Better user guidance on failure to capture non-const symbols in OpenQASM

[swernli]

Following up on #2417 (comment), it looks like the error clean up exposed an issue with how SemanticErrorKind::ExprMustBeConst is used in lower_ident_expr. The error is very helpful, but no longer gets filed because symbol resolution doesn't return actual symbols that are non-const. We need to review the interaction with try_get_existing_or_insert_err_symbol to ensure the assumptions are compatible with the lower_ident_expr.