Re-connect to Nomad if connection is lost

[lgfa29]

This PR fixes an issue where the Autoscaler would fail to re-connect to the Nomad server if the connection was dropped.

The main cause for this issue is that the policyIDsErrCh was not being consumed, and so the goroutine monitoring the list of policy IDs would block in case of an error.

I also uncovered some more subtle ways that the Autoscaler could fail to re-connect properly:

• policy handlers would exit in case of connection error and never be re-created.
• the policy list ID monitor would silently continue if the Autoscler re-connected to a Nomad cluster where the list of policies were different but the modify index had the same value.

To fix these problems I made a few assumptions:

• Policy handlers only exit when their context is closed. They will keep looping even if an error occurs.
• The policy manager is responsible for stopping handlers in case of an unrecoverable errors.
• The policy manager will re-run if it exists until its context is closed.

Closes #119

[jrasell]

Looks good.
One question about the change in the MonitorIDs func that I am interested to understand.

[cgbaker]

cursory examination is all i have bandwidth for, but this looks good