Fix scalers leaking

[holyketzer]

We are using Keda with RabbitMQ, after 30 days running of keda-operator pod we have more than 1000 open connections in RabbitMQ console. It wasn't our app or something else, we checked it by restarting step by step. Connections were dropped only after keda-operator restart.

Fixes:

In case of error GetDeploymentScalers and getJobScalers returns acquired scalers, but callers of these methods doesn't use them or close them. So it doesn't make sense to return scalers at all in case of error, and it should close all already opened scalers to avoid leaking.

[msftclas]

All CLA requirements met.

[ahmelsayed]

Good catch. Thanks @holyketzer for the fix! There is one minor comment about fixing the existing error message as well that would be good to fix in this change as well.
Other than that, the change LGTM

[holyketzer]

I renamed scalers to scalersRes because there was a name clash between variable scalers which overwrites namespace (or how it's called in Go) scalers and I wasn't able to use scalers.Scaler struct inside function with error:

pkg/handler/scale_handler.go:257:20: scalers.Scaler undefined 
(type []scalers.Scaler has no field or method Scaler)

[zroubalik]

Good catch, thank you!