I assume this is due to the bolts that mount the caliper to the fork. Do note that if the brake caliper is behind the fork, the force pushes the brake caliper slightly upwards and towards the fork.
If the brake caliper is in front of the fork, the force pushes the brake caliper slightly downwards and away from the fork.
The force component pushing the brake caliper away from the fork is the problematic one. For example, a 110kg rider + 20kg e-bike + 15kg cargo means 145kg has to be stopped. If braking at 0.6 g, that's 853 Newtons of braking force.
The wheel rolling diameter is 677 mm, whereas the rotor diameter is 160 mm. Thus the force at the caliper is 677/160 = 4.23 times the force at the tire. That's 3610 Newtons of force either pushing the caliper towards the fork or pulling the caliper away from the fork. Your choice -- I will choose that force to push the caliper towards the fork and not pull the caliper away from the fork. My choice means the caliper has to be behind the fork, not ahead of the fork.
If the caliper mounting bolts are tightened to the full recommended torque, they probably can withstand this 3610 Newtons of force. Tightening to the recommended torque should ensure about 5000 Newton clamping force. But the problem is, brakes are a safety critical item, and bolts that are working so close to their limit even if correctly torqued, and likely to be torqued by amateur mechanics, would be very likely to fail in case someone puts only 3000 Newtons of clamping force to the bolt.
Also, torque is actually a poor indication of bolt preload. If you torque to let's say 5 Newton meters, the preload could be anywhere between 4000 Newtons and 6000 Newtons depending on the quality of the threads, the type of grease, the amount of grease, etc. It isn't as simple as saying that 5 Newton meters for a M5 bolt equals 5000 Newtons of preload.
Remember how a bolted joint works: if the piece attached with a bolt to a plate is pulled, the bolt load does not change. Instead, what changes is the load between the piece and the plate. However, that only works to the point where the load between the piece and the plate is zero. At that point, the force begins to increase the load at the bolt. Quite soon the bolt fails, and if it doesn't fail by separating in two pieces, it's possible the movement at the joint could cause the bolt to self-unscrew.
However, if you push a piece towards the plate in which it has been attached, in every case the load fully goes via the interface of the piece and the plate. The bolt preload is never affected, no matter if the bolt is loose or tight.
Now, if the fork extended below the dropout, then you could attach the caliper at the extending portion in front of the fork, so the force would then not pull the caliper away from the fork but rather push it towards the extended portion of the fork. However, such an extended fork is (1) heavy, and (2) can't have dropout slots that are open towards the bottom but rather must use thru-axle, and thru-axle anyway solves the QR unscrewing issue with disc brakes so such an extended fork doesn't make sense. It would also look ridiculous.