HostnameEndpoint should expedite the next connection attempt after a previous attempt fails fast

[Nathaniel J. Smith]

Currently, Twisted's happy eyeballs code works by starting a new connection attempt, and then if that hasn't succeeded after 300 ms it starts a new one, repeat.

This is inefficient in the case where a connection attempt fails fast: if a connection attempt fails after, say, 10 ms, then we should start the next attempt immediately rather than sitting around doing nothing for 290 ms like the current code does.

This affects both the current HostnameEndpoint and the current version of Glyph's rewrite [1].

[1] ​https://github.com/twisted/twisted/compare/trunk...glyph:statemachine-hostnameendpoint