Test ECN on a path dropping all marked packets

[mxinden]

neqo-transport tests whether a path is ECN capable by marking the first 10 packets:

neqo/neqo-transport/src/ecn.rs

Lines 14 to 15 in 6664452

	/// The number of packets to use for testing a path for ECN capability.
	pub const ECN_TEST_COUNT: usize = 10;

On a path that drops all ECN marked packets, this could delay connection establishment significantly.

We should test whether this is the case.

Especially relevant, since https://phabricator.services.mozilla.com/D212959 would enable ECN on Firefox.

---

Related discussion on Slack quicdev.

@marten-seemann:

quic-go only starts ECN validation in 1-RTT. I see little value in the additional complexity of having to deal with correctly accounting for packets that never get acknowledged due to the packet number space being discarded beforehand in my ECN validation logic. Also, it makes the code 100% resilient against ECN blackholes, which is one less thing to worry about when debugging connection failures.

See also quic-go implementation: https://github.com/quic-go/quic-go/blob/master/internal/ackhandler/ecn.go