Use GCP selflink for tracking instances throughout

[Nashluffy]

When the compute API is first queried, the self link is used to form a resource ID that is put in the cache. After formatting, that looks like

projects/mock-project/zones/europe-west1-c/instances/fake-resource

but when creation events are received, the resource name uses the project ID instead of the number, which looks like

projects/12345678/zones/europe-west1-c/instances/fake-resource

This leads to two types of resource IDs being stored in the cache. To fix this, compute resource IDs are stored in the form

projects/mock-project/zones/europe-west1-c/instances/fake-resource

which is formed from the self link.

Additionally, preemption events do not contain a self link, but the format of the resource ID matches what we stored in the cache.

This also fixes a small bug where an instance would be tracked even if parsing the pubsub message failed