refactor: improve conn event mgr perf
Let's spend some time digging through profiles together tomorrow. We'll likely need to refactor this, but for now, I think we're just falling behind.
The machine that's having trouble has 10K peers and we have several loops that perform one operation per CID per peer. That means we're doing 10K operations per cancel. Given wantlists of 256 CIDs, that's 2M operations per wantlist.
This repository has been moved to https://github.com/ipfs/go-libipfs. There is not an easy way to transfer PRs, so if you would like to continue with this PR then please re-open it in the new repository and link to this PR.