If we are talking about a "prototype", a platform for evolving software, were you never know what inputs can end up being useful; then sure - you can add as many as possible for later use.
If we talk about a more mature system, well past the concept stage, then stripping away those without firm purpose going forward should make sense.
According to data from Tesla themselves, they still seem to be improving:
Not sure how they compare to others.
Well, the camera shows us the train and train crossing, the image recognition doesn't seem to do a good job. Yes, software relying on LiDAR could solve this, but improved image recognition can also solve it.
At the end of the day there isn't one definitive solution, there's plenty of pathways.