Change normalization of loss function

[facusapienza21]

The losses right now allow for a scalling of the final loss:

# Scale loss function
ℓ_scale = if lossType.scale_loss
  ℓ / mean(Vx_ref[mask].^2 .+ Vy_ref[mask].^2)^0.5
else
  ℓ
end

However, this scalling (probably cause of legacy code) does not properly scale the loss to be dimensionless, and it is rather arbitrary at this point. I will change the default behaviour to be no scalling and provide a few scalling options that make more sence (for example, remove the ^0.5 in the denominator, since this does not cancels the scale of the loss).