fix(wio-tracker-l2): stop the I2S clock between sounds; GPS and TCXO fixes

The speaker's I2S channel was only disabled when idle, which leaves MCLK
running. Its fractional divider (160 MHz / 39 1/16 = 4.096 MHz) makes a
comb of lines every 256 kHz: one at 869.632 MHz, in the EU narrow mesh
channel, lifted the LoRa noise floor from ~-102 to ~-85 dBm, and others
land near GPS L1 and BeiDou B1. The channel is now deleted and its pins
held low after each sound, and made again for the next one. Measured:
noise floor -102 dBm, outdoor first fix 87 s (was 620 s and more).

Also: TCXO at 3.0 V (the L2's part, as Meshtastic has it); GLONASS
enabled on the L76K after each power-up; begin() on a running receiver
no longer resets it (a reset threw away the acquisition in progress).

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

variants/wio-tracker-l2/target.h

@@ -16,15 +16,21 @@
 
 // The L76K's power and reset lines are on the IO expander, out of the NMEA
 // provider's reach: begin() / stop() / reset() drive them here, so GPS off
-// (Settings, duty cycling) really powers the receiver down.
+// (Settings, duty cycling) really powers the receiver down. begin() on a
+// running receiver does nothing: a reset there (the app setting "gps" again)
+// threw away the acquisition in progress.
 class L2GpsProvider : public MicroNMEALocationProvider {
   bool _uart_off = false;
+  bool _started = false;       // begin() ran (the boot one does the reset)
+  bool _cfg_pending = false;   // constellations to set once the module talks
+  uint32_t _cfg_rx = 0;
 public:
   using MicroNMEALocationProvider::MicroNMEALocationProvider;
   void begin() override;
   void stop() override;
   void reset() override;
   bool isEnabled() override;
+  void loop() override;
 };
 
 extern WioTrackerL2Board board;