Fix mcm_machine_set_failed_machines metric to more accurately reflect current state

[gagan16k]

What this PR does / why we need it:
This PR adds changes such that a gauge metric scraped on prometheus gets cleared and re-populated on every poll, so that it reflects the latest state of the reference slice *machineSet.Status.FailedMachines.
The logic to set this variable has also been updated so that it contains the current set of machines in the Failed state.

Which issue(s) this PR fixes:
Fixes #476

Special notes for your reviewer:
IT for mcm-provider-aws passed

Testing was done by setting the machineCreationTimeout to a value lesser than the time it takes for a machine to be provisioned and move into the Running state (1m). The machine moves to the Failed state, when it can be checked against a prometheus graph to verify if it was recorded accurately

Some changes were required to ensure a Failed machine stays failed, like an artificial delay in the goroutine marking a machine for deletion

Stacked graph reflecting number of failed machines accurately matched number in the cluster

Release note:

Fixed metric `mcm_machine_set_failed_machines` and underlying variable `*machineSet.Status.FailedMachines` so that they reflect the current state of machines

[CLAassistant]

All committers have signed the CLA.

[takoverflow]

LGTM
Please update the release notes (you can check other PRs for reference)

[aaronfern]

Looks good to me!
Just a small nit