hive gossip improvements

[acud]

Right now there's a few limits regarding how and what we gossip on hive.

• randomSubset in kademlia - this was set when the amount of peers gossiped was so large that it was consuming way too much bandwidth. We should set this to be a more reasonable higher value (but also have a cap in a way that closer bins can still get gossiped correctly and fully)
• add write coalescing in hive - when frequent connections occur, it is very easy that single peer gossip messages get sent. This is wasteful. We should emit a message every second or so when gossiping single peers (if the message is big it can be dispatched immediately). Single peer gossips wait

[sbackend123]

I have additional improvement:

Currently, in distant bins (below storageRadius), we pick 2 random connected peers per bin (BroadcastBinSize = 2).

This ignores two aspects:

• A bin may have 5 peers or 20. Sharing with 2 means ~40% coverage in the first case and only 10% in the second.
• “Distant” bins can sit next to the neighborhood boundary or be as far as possible.

I think, it makes more sense to follow: the closer the bin to the neighborhood, the higher the coverage.

So I suggest , in short: we replace flat “BroadcastBinSize per bin” with a total fanout_cap on distant bins only, weighted toward bins closer to storageRadius, with random picks within each bin’s quota; neighborhood gossip is unchanged.

Example: with 12 active (not empty) shallow bins and BroadcastBinSize = 2 , so today ~12 × 2 = 24 recipients, which will be shared info bout new peer with.

With fanout_cap = 36 (same total as flat 3 per bin):
bin 11 — up to 6
bin 10 — up to 5
…
bin 0 — 1
Total ≤ 36.

We get faster, denser discovery near the neighborhood boundary and in “almost-neighbor” bins with same traffic.
@acud @gacevicljubisa @janos