Stop the slow test packages waiting (closes #263)

- Tests hash passwords at bcrypt.MinCost, set through the
  PasswordCost field of db.Params.
- Test IRC servers check the queue only on broker notification;
  one test keeps the 100 ms fallback poll.
- The stop-timeout test shortens the timeout through a test-only
  setter.
- service and handlers tests build config.Config directly instead
  of through config.New, so they run in parallel: service had run
  serially, and handlers started its servers one at a time.
- Repeated race trials run as parallel subtests.
- Fixed sleeps become waits on the broker.
- A new test covers the relay's failed queue read, which only the
  100 ms poll had reached.

Model: opus-5-5

internal/ircserver/server.go

@@ -53,6 +53,11 @@ type Server struct {
 	pingInterval time.Duration
 	pongDeadline time.Duration
 
+	// pollInterval and stopTimeout are the constants of the
+	// same names, except in tests, which change them.
+	pollInterval time.Duration
+	stopTimeout  time.Duration
+
 	// resolveHost is the function of the same name, except in
 	// tests, which replace it with one that blocks.
 	resolveHost func(ctx context.Context, addr string) string
@@ -79,6 +84,8 @@ func New(
 
 		pingInterval: pingInterval,
 		pongDeadline: pongDeadline,
+		pollInterval: pollInterval,
+		stopTimeout:  stopTimeout,
 		resolveHost:  resolveHost,
 	}
 
@@ -154,10 +161,10 @@ func (s *Server) stop() {
 
 	select {
 	case <-done:
-	case <-time.After(stopTimeout):
+	case <-time.After(s.stopTimeout):
 		s.log.Error(
 			"irc connections not cleaned up before stop timeout",
-			"timeout", stopTimeout,
+			"timeout", s.stopTimeout,
 		)
 	}
 }
@@ -181,7 +188,7 @@ func (s *Server) acceptLoop(ctx context.Context) {
 
 		client := newConn(
 			tcpConn, s.log,
-			s.database, s.brk, s.cfg, s.svc,
+			s.database, s.brk, s.cfg, s.svc, s.pollInterval,
 		)
 
 		s.mu.Lock()