UserInterface: poll the network stream on the 60 Hz update clock

40250 runs CPythonApplication::Process every 16/17 ms and only skips the
render when late, so GamePhase gets its three-packets-a-poll budget 60
times a second. The host polled once per rendered frame: at 30 fps idle
(or a slow phone) mob move packets queued behind character add/del
bursts for up to a second and mobs stopped, then jumped. Poll once per
elapsed 60 Hz tick. Real-server A/B: move push gap p50 138-178 -> 70-92
ms, p90 580-785 -> 358-403 ms.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

src/platform/UserInterface/PythonApplication.cpp

@@ -330,7 +330,23 @@ bool CPythonApplication::Process()
 	// Network I/O
 	{
 	MtPerf::CScope kPerf(MtPerf::SECTION_NETWORK);
-	CPythonNetworkStream::Instance().Process();
+	// PORT: 40250 runs Process() on its 60 Hz update clock and only skips the render pass when late
+	// (s_bFrameSkip), so the stream is polled every 16 ms whatever the render rate, and GamePhase
+	// handles three packets a poll while less than 8 KB is pending. The host runs one Process() per
+	// rendered frame (30 fps when idle, less on a busy phone), which halves that budget or worse: the
+	// character add/del bursts of a mob field hold the move packets back for up to a second and the
+	// mobs stop, then jump. Poll once per elapsed 60 Hz tick, as 40250 does.
+	{
+		static DWORD s_dwNextNetworkPoll = 0;
+		const DWORD dwNow = ELTimer_GetMSec();
+		if (!s_dwNextNetworkPoll || int(dwNow - s_dwNextNetworkPoll) > 500)
+			s_dwNextNetworkPoll = dwNow;
+		do
+		{
+			CPythonNetworkStream::Instance().Process();
+			s_dwNextNetworkPoll += 16 + (s_dwNextNetworkPoll & 1);
+		} while (int(dwNow - s_dwNextNetworkPoll) >= 0);
+	}
 	// PORT: m_kGuildMarkUploader/m_kGuildMarkDownloader.Process() follow here in 40250; the guild mark
 	// transfer is not ported yet (GuildMarkDownloader.cpp/GuildMarkUploader.cpp).
 	CAccountConnector::Instance().Process();