Improve how we detect database port is free

[infomiho]

When you run nc -l 5432 on the dev-database port:

$ lsof -nP -i :5432 | grep LISTEN
nc      67588 ilakovac    3u  IPv4 0x2b284e848f681f22      0t0  TCP *:5432 (LISTEN)
nc      69683 ilakovac    3u  IPv4 0x8350f3e3a67bb38f      0t0  TCP *:5432 (LISTEN)

wasp doctor claims the port is free:

$ wasp doctor
[✓] Dev database port (5432): free

and starting the dev database dies with a raw runtime error instead of the intended friendly message:

$ wasp start db
wasp-cli: Network.Socket.connect: <socket: 147>: timeout (Operation timed out)
HasCallStack backtrace:
  bracket, called at lib/System/IO/Utf8.hs:154:24 in wth-tf8-1.1.0.0-1dd5724a:System.IO.Utf8

exit=1

Cause: two stacked probe flaws around port 5432:

1. Doctor's only probe is checkIfPortIsInUse (src/Wasp/Util/Network/Socket.hs:37), which binds 127.0.0.1:PORT with SO_REUSEADDR set - that succeeds under a wildcard (*:5432) listener, so the port reads as free. start db knows about this hole and adds a second probe (checkIfPortIsAcceptingConnections, comment at cli/src/Wasp/Cli/Command/Start/Db.hs:146)...
2. ...but that probe connects, and any exception other than ECONNREFUSED is rethrown (Socket.hs:24). Against nc -l the connect times out; the resulting IOException escapes both the requirement wrapper and Main.hs's ErrorCall-only handler, producing the crash above. The clean [Error] Port already in use path only triggers for ports that refuse connections outright.