Refactor SlippiNetplayClient to enable attempting connections to multiple endpoints per player

[jmlee337]

(at this time, the only possible endpoints are external and local)

Because this rewrites much of the connection logic, I tested this in the following cases

• "normal" singles
• singles via LAN
• singles vs symmetric NAT
• doubles (2 LAN players) vs double symmetric NAT (2 LAN players)

In the immediate term this will fix connections for the case where

• multiple players are connected to the same VPN AND
• the selected local address is not correct (due to the VPN client not appearing as a network device)

Before #387 is merged this will also fix connections for the case where

• multiple players are connected to the same LAN AND
• the selected local address is not correct (due to the behavior fixed by Determine the actual used local address with ENET socket #387) AND
• the LAN gateway supports hairpinning

In the future there is the possibility of using this functionality to

• improve connections for players who are behind the same CGNAT/double NAT
• enable integrated relay servers which would be deployable by anyone, replace community VPNs, and be easier for both users and administrators

[JLaferri]

Oh boy. Yeah this changes a lot of critical things. Some of the stuff that got changed were things that I added because there were issues with people disconnecting from each other 30 seconds into a match, for example.

Would be good to do some decent testing and also have a decent revert plan if something breaks.

[jmlee337]

I'll have to think more about a testing plan. The main problem I'm thinking of is that this PR alone doesn't offer any improved functionality to entice testers (in 99.9% of cases).

In the meantime can you tell me about the 30 second disconnect? What parts of the original code address that? I can at least try to make sure there's no regressions.