Make the number of events buffered to/from tasks configurable

[tomaka]

cc paritytech/substrate#6009

Hints seem to indicate that we have some CPU issues in Substrate from the low number of events that can be buffered. This PR makes this limit configurable.

Additionally, this PR increases these values: the channel size from tasks to the swarm is now 32 (instead of 1!), and the channel size from the swarm to each task is now 7 (so 8 in practice).

When reviewing, keep in mind that channels reserve an extra element for each active sender, as documented: https://docs.rs/futures/0.3.5/futures/channel/mpsc/fn.channel.html

(I also slightly fixed the CHANGELOG to be alphabetically-ordered)

[romanb]

I left only naming and documentation suggestions that are motivated as follows:

• Internally, I previously established the terminology of "task events" vs "task commands", rather than speaking of "events from" or "events to" tasks, respectively, so that naming in the pool and manager settings would be more consistent.
• On the public API, I don't think it is beneficial to talk about (background) tasks. Instead, a user of libp2p-core knows that notify_handler is used to send events to the connection handlers and that he receives connection events by polling the network. Hence buffer sizes that relate to notify_handler and "connection events" are directly meaningful, especially with the added documentation about the effect of a full buffer.

[mxinden]

For what my review is worth, this looks good to me.