core: etcs: for LOAs, compute EBD foot and EBD curve more accurately

[Erashin]

In ETCSBrakingCurves.compûteBrakingCurves, in the case of an EBD with a non null target speed, we are doing a couple of approximations:

• once we intersect with targetSpeed + dvEbi, we just shift the envelope to the right.
• if the LOA is close to the start of the path, the EBD could just reach the start of the path before reaching targetSpeed + dvEbi: in this case we don't shift it, which is another approximation

This creates a few small problems:

• we're not using the correct gradients for the calculation (by an offset of around ~50m most of the time, since dvEbiMax = 7.5km/h, it's not really high) -> not the biggest problem, since the offset tends to be small, but still not the best.
• if the EBD hits the start of the path, and we shift it, the curve won't start at position = 0.0. Potentially (almost never), that'll mean that the indication curve could start after 0.0. -> in this case, we'll use const gamme to brake from 0.0 till EBD curve, and we won't even see it in the MSP, because we'll skip this entire part with the time it takes for the train to accelerate and reach the MSP. This changes, however, if we make the train start with a beginSpeed (which we almost never do)..

We should really compute the correct curve, i.e., start from the point (LOAPosition, LOASpeed + dvEbi), decelerate to the left until MRSP or position = 0.0, then decelerate to the right until the point where we reache LOASpeed.