Pre-existing bug: in the Linux `ReadFile.doReadLoop` path, `this.read_off` is initialized to 0 and never incremented during the loop, so `remainingBuffer()` always caps each individual read at `max_length` bytes rather than `max_length - already_read` bytes. When data arrives in partial chunks, the accumulated buffer can exceed `max_length` by up to `max_length-1` bytes. The PR adds a test claiming to verify the N-byte cap, but the test writes all bytes atomically so the OS delivers exactly N by