notifications: Investigate possible memory leak with pending outbound substream tracking

[lexnv]

The metrics implementation exposed an abnormal behaviour in our state tracking:

• metrics: Expose litep2p metrics in an agnostic manner #294

It looks like the following data structure grows over time:

litep2p/src/protocol/notification/mod.rs

Lines 261 to 262 in 28d42c6

	/// Pending outboudn substreams.
	pending_outbound: HashMap<SubstreamId, PeerId>,

Elements are added to the hashmap on:

• litep2p/src/protocol/notification/mod.rs

  Line 919 in 28d42c6

  async fn on_open_substream(&mut self, peer: PeerId) -> crate::Result<()> {

• litep2p/src/protocol/notification/mod.rs

  Line 1093 in 28d42c6

  async fn on_validation_result(

Elements are removed from the hashmap on:

• litep2p/src/protocol/notification/mod.rs

  Line 514 in 28d42c6

  async fn on_outbound_substream(

• litep2p/src/protocol/notification/mod.rs

  Line 816 in 28d42c6

  async fn on_substream_open_failure(

Possible causes

It looks like we should remove the state on the connection closed as well:

litep2p/src/protocol/notification/mod.rs

Line 445 in 28d42c6

OutboundState::OutboundInitiated { .. }