Weird behavior when working with team ball

[Flova]

Summary

In https://data.bit-bots.de/ROSbags/german_open_2024/ID_jack_2024-04-20T13%3A36%3A16/ID_jack_2024-04-20T13%3A36%3A16_0.mcap at 2024-04-20 1:40:16.618 PM CEST we go to block a ball that is behind the goal. But the ball was only detected shorty with a bad covariance behind the goal. We somehow still block the ball behind the goal after we detected the correct ball (but with a bad covariance). Somehow the ball seen decision passes because of the team communication, but we use an outdated bad detection from ourselves instead of the team ball.

Example image. The ball has a very bad covariance (the purple blob is larger than the screen). Yet we still say that we saw the ball (I think due to team comm).

Previously (> 50s earlier) there way a "bad" ball filter prediction behind the goal (possibly due to a velocity over estimation. See #424). It seems like we stick with this one and it does not time out due to the team ball detections, but we somehow prefer it over the team ball when we go to the block position.