syncer: minimize number and size of requests in ForkFinder

[fasmat]

Description

The ForkFinder in the syncer package at the moment requests the aggregated hashes of multiple layers at once, up to MaxHashesInReq (100) to find where the local state started differing from a peers state. This is can be made more efficient by instead using a binary search:

The local node should request the aggregated hash in the middle between the From and To and then narrow the range until the first divergent layer is found. This limits the number of requests and hashes to log2(To - From) and removes the need for checks on the client and server side limiting the amount of layers that are requested at once.

Affected code

• ForkFinder in the syncer package
• handler in the fetch package
• sql/layers

For more info see discussions in PR https://github.com/spacemeshos/go-spacemesh/pull/4480/

[fasmat]

As an alternative the ForkFinder can also just request a stream of layer hashes covering the range of From and To, compare hashes and stop the stream as soon as the first divergent layer is found.

[fasmat]

Before changing the implementation we should performance test the current solution and answer the following questions:

• how large is the usual search range the fork finder has to cover?
• where within that range does the divergence usually happen? Is it uniformly distributed or is there a bias towards older or newer layers?

The larger the usual search range is the better a binary search should perform compared to a streaming solution, unless it is more likely to find the forking layer close to the beginning (or end) of the search range.

Additionally it needs to be tested if fewer large request or more smaller requests are more efficient (requesting up to 100 at once vs. requesting only 1 layer).