Optimize subgraph synced check

[leoyvens]

Every time we process a block, we check if the subgraph has reached the chain head and can be considered synced. Right now this requires a roundtrip to the DB to check the chain head, here:

graph-node/core/src/subgraph/runner.rs

Line 753 in fe96516

self.inputs.chain.chain_store().chain_head_ptr().await?,

This overhead is unnecessary, since if a subgraph checks the chain head once and sees that it is very far, there is no need to check again on every block processed.

[lutter]

In a similar vein, there is a check store.is_deployment_synced() in the runner to see if the subgraph is synced:
https://github.com/graphprotocol/graph-node/blob/master/core/src/subgraph/runner.rs#L590

The runner should know whether a subgraph is synced and only ask the store for that when it starts a subgraph; after that, it already is in control of the synced status and keep that in memory.