fix(esp32s3): bound the SerialConsole idle sleep on hardware USB CDC (#11901)

HWCDC::isPlugged() is a SOF watchdog that reads false transiently while USB
is connected and working. runOnce() answered that with a 20 s sleep, and
nothing wakes the thread on RX, so host traffic sat in the CDC RX ring and
reached the API as a burst. Cap the sleep at 250 ms, the rate readStream()
already idles at.

IS_USB_SERIAL only tested ARDUINO_USB_CDC_ON_BOOT, so ARDUINO_USB_MODE=0
boards ran the same check against a USB-Serial/JTAG peripheral that is not
attached to the PHY and never sees a SOF. Gate the check on IS_USB_HWCDC.

Measured on tlora-t3s3-v1, 900 s of 1 Hz ToRadio/FromRadio round trips:
before 12 stalls, rtt_max 19.96 s, console asleep 26.4% of wall time.
After 0 stalls, rtt_max 0.147 s, p50 unchanged at 0.028 s.

Fixes #11864

src/SerialConsole.cpp

@@ -16,6 +16,11 @@
 #ifdef SERIAL_HAS_ON_RECEIVE
 #undef SERIAL_HAS_ON_RECEIVE
 #endif
+// Port is HWCDC only in hardware USB-Serial/JTAG mode. With ARDUINO_USB_MODE=0 it is TinyUSB
+// USBCDC, the PHY is routed away from the USJ peripheral, and isPlugged() would never see a SOF.
+#if defined(ARDUINO_USB_MODE) && ARDUINO_USB_MODE
+#define IS_USB_HWCDC
+#endif
 #include "HWCDC.h"
 #endif
 
@@ -130,8 +135,10 @@ int32_t SerialConsole::runOnce()
     if (hasPendingOutput())
         return delay < 25 ? delay : 25; // 0 continues a budget slice; else short-poll TX drain
     return Port.available() ? delay : INT32_MAX;
-#elif defined(IS_USB_SERIAL)
-    return HWCDC::isPlugged() ? delay : (1000 * 20);
+#elif defined(IS_USB_HWCDC)
+    // isPlugged() is a SOF watchdog that flaps false while USB is fine (#11864), and nothing wakes
+    // this thread on RX, so cap the idle sleep at the rate readStream() already idles at.
+    return HWCDC::isPlugged() ? delay : 250;
 #else
     return delay;
 #endif