Fix typo in fast templates test

[alejoe91]

@DradeAW there was a typo in the tests, but still fast and normal templates give the same stds down to the 4th decimal (which I think it's ok)

[DradeAW]

Yeah, I'm getting -5943.21 sometimes, so I don't think it's a rounding issue ^^

I'll go digging to understand what's going on!

[DradeAW]

This is good for me,

Thanks Alessio :)

[samuelgarcia]

We need a copy only if computing STD otherwise the copy is useless.

[DradeAW]

I don't like this,
Because now, if return_std=False, then waveforms_sum is changed (pointer problem)

Of course it's not used in this case, but I think it's bad practice (what if someone implements something in 8 months from now? This is a potential bug)

[alejoe91]

I don't like this, Because now, if return_std=False, then waveforms_sum is changed (pointer problem)

Of course it's not used in this case, but I think it's bad practice (what if someone implements something in 8 months from now? This is a potential bug)

This was a request from @samuelgarcia, since copying the buffer can be large for many units...

[samuelgarcia]

memory allocation is not a detail. we should carrefully have this in mind.
averaging should need one allocation and divide in place (like it was the case).
the preprocessing need a big change mainly because of this for instance.

[DradeAW]

I agree that memory allocation is not a detail and not something to take likely,

But having two variables (representing completely different things) pointing to the same address seems dubious to me ...