perf: move per-packet datapath logging to DEBUG

[kedimuzafer]

Problem

A handful of log statements on the WireGuard/DISCO datapath are at ESP_LOGI, and two are plain printf(). On a default build they therefore fire once per packet.

At 115200 baud a ~70-character line takes ~6 ms to write, and the UART write blocks the task that issues it. ml_net_io.c's UDP RX: line is emitted from the socket receive task before the packet is even queued, so the cost lands directly in the receive path.

Measurement

ESP32-S3 (8 MB PSRAM) running MicroLink as a Tailscale SOCKS5 proxy, peer on the same LAN, direct path:

	before	after
sustained packet rate before loss (1000 B ICMP)	~20 pps, then 18% loss at 50 pps	100 pps, 0% loss
tunnel RTT	20–235 ms, mdev 58 ms	6.2 ms, mdev 0.8 ms

Before the change, any TCP transfer larger than a few tens of KB through the tunnel stalled, and the node stopped answering ICMP for the duration — the UART was holding the receive task while the lwIP UDP mailbox (CONFIG_LWIP_UDP_RECVMBOX_SIZE, 6 by default) overflowed behind it.

Change

Demoted to ESP_LOGD / WG_DEBUG so they compile out at the default log level and remain available by raising it:

• ml_net_io.c — UDP RX: ..., once per received packet
• wireguardif.c — [WG_RX] type=..., once per received packet, a raw printf
• ml_wg_mgr.c — DISCO RX/PING/PONG, WG RX/WG UDP TX, the WG INIT hex dump, and the wireguardif_periodic / disco_periodic_probes timing lines
• ml_derp.c — DERP RecvPacket/SendPacket/PeerGone, HEARTBEAT

Also removes the [TAI64N] uptime printf in wireguard_tai64n_now(). It is rate-limited to every 5 s, but it sits on the handshake path and reports nothing actionable.

Nothing changes except log verbosity. Everything demoted is still reachable with esp_log_level_set() or a higher CONFIG_LOG_MAXIMUM_LEVEL.