Fix W5100S socket exhaustion blocking MQTT and additional TCP clients

[PhilipLykov]

Summary

Fix W5100S socket exhaustion on RAK4631 Ethernet gateways that prevents MQTT connections, DHCP lease renewal, and additional TCP API clients.

The W5100S chip has only 4 hardware sockets. With NTP and Syslog enabled, all 4 sockets were permanently consumed:

Socket	Service	Lifecycle
0	NTP UDP	Permanent (held from boot)
1	Syslog UDP	Permanent (held from boot)
2	TCP API Server (LISTEN)	Permanent
3	TCP API Client (e.g. MeshMonitor)	While connected

This left zero sockets for MQTT, DHCP lease renewal, or NTP updates — causing MQTT to silently fail with no packets sent on the wire.

Changes

• NTP on-demand (ethClient.cpp): Remove permanent timeClient.begin() at startup. NTPClient::update() auto-initializes when _udpSetup is false. Add timeClient.end() after each query to release the socket immediately.
• Syslog on-demand (DebugConfiguration.cpp): Remove socket allocation from Syslog::enable(). Open and close the UDP socket in _sendLog() around each message send. Syslog overhead is ~320μs per message for socket open/close, well under 1% CPU at typical log rates.
• MQTT validation socket leak (MQTT.cpp): Fix isValidConfig() where a successful test connection was never closed — PubSubClient::~PubSubClient() only frees buffers, it does not call disconnect() or _client->stop(). The leaked socket in ESTABLISHED state permanently consumed a W5100S slot.

After fix

Socket	Service	Lifecycle
0	TCP API Server (LISTEN)	Permanent
1	TCP API Client	While connected
2	MQTT Client	While connected
3	NTP / Syslog / DHCP (shared)	On-demand, brief

Test plan

• RAK4631 Ethernet gateway with syslog, NTP, MQTT, and API client all enabled simultaneously
• Verify MQTT connects successfully while MeshMonitor is connected
• Verify syslog messages still appear on the syslog server
• Verify NTP time sync works after the change
• Verify DHCP lease renewal works (check over 24+ hours)
• Verify no regressions on ESP32 WiFi builds (syslog change affects all HAS_NETWORKING platforms)

[robekl]

args is reused without va_copy(); reusing a traversed va_list is undefined behavior and can produce corrupted logs or crashes

[robekl]

return value of endPacket() is ignored; true is returned unconditionally. On embedded Ethernet stacks, endPacket() is the point where TX failure is reported; this currently reports success even when the datagram send fails.

[robekl]

Socket cleanup is conditional on client == nullptr; when a non-null client is passed, successful validation can still leave the connection open. Since this function is a validator, that side effect can leak scarce sockets on Ethernet-class devices. Consider ensuring cleanup for all validation paths (or explicitly documenting and renaming if persistent connection is intended).

[thebentern]

@robekl please stop using LLMs to review random PRs. We have CoPilot for this. Additionally the reviews are also commenting on existing unchanged code. This is your one and only warning.

[PhilipLykov]

@robekl Thanks for the review.

1. va_copy() — This is pre-existing code in vlogf() that we did not modify. Valid observation, but out of scope for this PR.

2. endPacket() return value — Also pre-existing behavior in _sendLog(). Our change only added the begin()/stop() wrapping around it.

3. Conditional disconnect() — Intentional. When client != nullptr, a caller-provided mock is in use (unit tests assert client.connected_ stays true after isValidConfig()). In production client is always nullptr, so disconnect() always runs.

[PhilipLykov]

Resolved the merge conflict in MQTT.cpp — upstream rewrote isValidConfig() to use a local MQTTClient testClient with an explicit testClient.stop(), which inherently fixes the socket leak our PR addressed in that function. Took the upstream version.

The other two fixes in this PR are still needed and unaffected by upstream:

• NTP on-demand (ethClient.cpp): timeClient.end() after each NTP query to release the UDP socket — upstream still holds it permanently from boot.
• Syslog on-demand (DebugConfiguration.cpp): Open/close the UDP socket in _sendLog() per message — upstream still holds it permanently from enable().

Without these two changes, the W5100S still exhausts all 4 hardware sockets (NTP UDP + Syslog UDP + TCP listener + TCP client), leaving none for MQTT, DHCP renewal, or additional connections.