Use mili scale to support more precision

[JorTurFer]

Proposal

K8s supports using different units to bring more precision to the values but right now, all the scalers return an integer quantity in the default unit. This can produce data lost in some cases.

Use-Case

Imagine that you have one metric which is a rate from 0 to 1, with the current approach you can't use KEDA for scaling it without any other different solution but using mili in the target and the value you could do it.

Anything else?

To unify the scaler I'd just return all the metrics/targets in mili. With several scalers doesn't make sense because they are always integers (message count in a queue is an integer), but it will work well even that and unifies the approach across all the scalers