Frequent allocations setting context deadlines

[Stebalien]

[dirkmc]

The reason for the allocation is that we set a timeout each time we send a message.
I believe we're getting so many of them because when we send a broadcast want or a cancel to thousands of peers we set a timeout for each one.

One idea that occurs to me is to keep a pool of contexts with timeout, where the timeouts are close to each other.
For example for a broadcast to peer A, peer B and peer C, the message timeout deadline will be roughly the same for all three. So before setting the timeout they check the pool to see if there's a context with timeout that falls within eg 10ms of the deadline they need.

[Stebalien]

As far as I can tell, we only need this context in some special cases (e.g., when calling Connect). We don't do that every time. Can we create a context on demand?

[dirkmc]

That's true we only need it when calling connect and when opening a stream, but that should only happen once per peer. So actually I think we can avoid using WithDeadline altogether for most calls to SendMsg()