[Access] Refactor the collection indexing and syncing

[peterargue]

Expands on #8108

The original implementation had 2 issues discovered when we deployed it to live nodes:

1. When indexing from execution data, the syncer loads data faster than the indexer can write it back to storage. This caused the indexer's pending collection queue to overflow.
2. The syncer's request missing collections method first scanned all blocks from the last full block to the latest finalized to build a list of missing collections, then submitted them all at once to the network. When a node is very far behind, and has a slow disk, this process can take a very long time causing indexing to pause.

Addressed 1 by making the syncer block when indexing from execution data.
Addressed 2 by refactoring the syncer to process each block sequentially.

[peterargue]

removed this because we had redundant checks. if there are MissingCollsForBlockThreshold blocks with missing collections, then the last full block is at least that many blocks behind the last finalized. Updated to only check the difference between last finalized and last full

[zhangchiqing]

Changes looks good.

I have one concern for syncer's requestMissingCollections :

The syncing asks the requester to request missing collections, after a few seconds, the execution data indexer indexed the execution data, then the syncer is able to index the collections, and the indexer will move forward the last full block height. However, the requester didn't know about the fact that the collections have been received, so I think over time, it will build up a long list of zombie collections in the fetch list, we need to check how the requester deals with it, I'm afraid those zombie collections will affect the actual missing collection requests from being sent.

[peterargue]

that's a good point. the requester should eventually download the collections, but it is wasted effort. it may be worth adding a timeout for AN collection syncing