GitLab

Instead of copying these slices, we can just reuse them.

In case we're sending a _lot_ of wants:

* Prioritize cancels.
* Then targeted wants.
* Finally broadcast wants.

This makes it easier to tell where module boundaries are.

This removes one goroutine per peer which tends to be a pretty big deal. This
brings go-ipfs down from 5.5 to 4.5 goroutines per peer.

And expose a separate function for _just_ getting want-blocks. When the user
runs `ipfs bitswap wantlist`, they expect to see everything the node is
currently looking for.
Co-Authored-By: dirkmc <dirkmdev@gmail.com>

We use these outside of bitswap for testing.