jsonrpc: fix error when we abort batching due to timeout.

The read_json() call expects len_read to be the amount of *new*
data read.  If we call this back without resetting, it will parse
this much random junk in the buffer.

Fixes: #5766
Changelog-Fixed: Plugin: `autoclean` could misperform or get killed due to lightningd's invalid handling of JSON batching.

lightningd/jsonrpc.c

@@ -1109,6 +1109,8 @@ static struct io_plan *read_json(struct io_conn *conn,
 							  start_time),
 					 time_from_msec(250))) {
 				db_commit_transaction(jcon->ld->wallet->db);
+				/* Call us back, as if we read nothing new */
+				jcon->len_read = 0;
 				return io_always(conn, read_json, jcon);
 			}
 		}