Start metrics server first before instantiating the rest of ingester

[masih]

Admin server also exposes the debugging endpoints. In daemon, it is started after the ingester instantiation is finished. As a result if the code is slow to start we have no visibility into what's happening.

Refactor the code such that:

• debugging endpoints are accessible as soon as possible and do not depend on initialization of heavy weight components
• the node remains "unready" until all components are instantiated. This means the readiness check endpoint also exposed by the admin server should return non 200 until all components are initialised.