Fast messsage id cache

[twoeths]

Motivation

• Gossipsub is designed to have message redundancy, right now we have to calculate 3.97 times per message data in average, in the worse case we calculate 10/11 times per message data

Description

• Follow other clients to have a cheaper faster message id just for message de-duplication, so we have fast message id and canonical message id
• Key as the sha256 of the message (fast message id), value as the canonical message id => enhance SimpleTimeCache for this
• Instead of passing getMsgId() function to other classes (PeerScore, GossipTracer, MessageCache) in the constructor, just pass message id through method as we always have them in the consumer code which make it simpler (and Lighthouse does this!)
• Use the new getCanonicalMessageIdStr() function in all places instead of getMsgId() since we can't base on getMsgId() in lodestar, the message is removed very quickly from WeakMap
• getCanonicalMessageIdStr(): get from the newly created getCachedMsgId() first, then the fast message id cache, then getMsgId

Test result

• Message id is calculated 1.02 times per message, max is 2
• This uses the default cache time which is 30s, should we increase it? Lighthouse uses a duration of 33 slots https://github.com/sigp/lighthouse/blob/fff01b24ddedcd54486e374460855ca20d3dd232/beacon_node/eth2_libp2p/src/config.rs#L36 . We can change the seenTTL option anyway
• With this 30s cache, it takes 1.2MB (and 9556 cache entries) if I use subscribeAllSubnets flag in lodestar

[twoeths]

Lodestar getMsgId() should divide its implementation into getCachedMsgId() and computeMsgId()

[codecov-commenter]

Codecov Report

Merging #179 (7dd97e0) into master (1a442fe) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master     #179   +/-   ##
=======================================
  Coverage   76.58%   76.58%           
=======================================
  Files           2        2           
  Lines        1905     1905           
  Branches      140      141    +1     
=======================================
  Hits         1459     1459           
  Misses        446      446           

Impacted Files	Coverage Δ
test/utils/index.js	87.50% <ø> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 1a442fe...7dd97e0. Read the comment docs.

[dapplion]

This looks really nice! Thanks for looking into the memory costs 👍

Can you please open an issue to raise the timeout to 33 slots? Memory should be analyzed for that case too

[wemeetagain]

Maybe the fastMsgId should be passed into the constructor, like the msgId in the pubsub base class, so we don't need to add this dependency (maybe just dev dependency for testing).

[dapplion]

Looks good! Agree with Cayman suggestion to pull sha256 dependency

[wemeetagain]

I think the fast id feature should be optional. Likely many applications (including ipfs and existing apps) will not need this feature and shouldn't pay for the cost.
We can switch some of this logic based on if the fast id function is set or not. eg: if (this.fastIdFnStr) { ... } else { ... }