test(chaos): say why short_gap_collision is bounded by state

Follow-up comment to the previous commit, which made the collision
ordering explicit. The case table now records what the bound rests on
and what failed before it, so the next reader does not re-derive it
from the runner. No value changed.

Author: thomas-mangin

internal/chaos/inprocess/runner_test.go

@@ -269,12 +269,20 @@ func TestInProcessDisconnectReconnect(t *testing.T) {
 		minDisconnected int
 	}{
 		{
-			name:            "short_gap_collision",
-			reconnectDelay:  0,                // Collision mode: queue new conn before closing old.
-			duration:        15 * time.Second, // Session is ESTABLISHED when new conn arrives.
-			minEstablished:  1,                // Initial session only — collision rejects the new one.
-			maxEstablished:  1,                // RFC 4271 §6.8: ESTABLISHED state rejects incoming.
-			minDisconnected: 0,                // Disconnect events vary by timing.
+			name: "short_gap_collision",
+			// Collision mode. The runner delivers the new connection while the
+			// reactor reports the peer ESTABLISHED and closes the old one only
+			// after the reactor has refused the new one, so the count below is
+			// bounded by the session's state, not by how the Go scheduler
+			// happened to order the accept handler against the old connection's
+			// EOF. Before that ordering was made explicit, losing the race let
+			// the new connection establish a SECOND session and this case failed
+			// "2 is not less than or equal to 1" on a slow host.
+			reconnectDelay:  0,
+			duration:        15 * time.Second,
+			minEstablished:  1, // Initial session only — collision rejects the new one.
+			maxEstablished:  1, // RFC 4271 §6.8: ESTABLISHED state rejects incoming.
+			minDisconnected: 0, // Disconnect events vary by timing.
 		},
 		{
 			name:            "borderline_gap",