Remove reduntant error code store for no-interrupt case

[kurapov-peter]

No description provided.

[ienkovich]

Why is the error code store redundant? If dynamic interruption is disabled we still want to get error codes like overflow, etc.

[kurapov-peter]

What happens is we already store it with record_error_code inside the row function, then at the upper level we load the value to check if an error occurred, and if yes we store it back again. I discovered this while triaging the GPU hanging issue on a specific query. I wasn't sure that's always the case (even though we seem to always do it unconditionally, see createErrorCheckControlFlow), so I created this draft to check with CI.

[ienkovich]

I saw cases when row_func doesn't even get error_code but still can return a non-zero value. This happens when the number of matched rows exceeds max_matched value. In this case, we return -pos written later to error_code. I'm not sure if this is used in tests, but looks like a useful feature to interrupt and then continue execution from the last position.