…to OTG PHY

Root cause (TinyUSB hathach/tinyusb#2943, diagnosed by hathach; fix merged as
adafruit/circuitpython#9973): on the ESP32-S3 the USB D+/D- pins power up muxed
to the ROM USB-Serial-JTAG controller, and the app must switch the mux to the
USB-OTG (TinyUSB) PHY. As of ESP-IDF >= 5.2 (we build on 5.4.1) usb_new_phy()
no longer flips that mux reliably / early enough, so the host enumerates the
JTAG device (303a:1001 "USB JTAG/serial debug unit") and keeps it -- our TinyUSB
CDC+MIDI device (caf0:4009) never appears. Worst on Windows / direct-connect; a
powered hub masks it by changing reset timing.

That also explains why the earlier experiment in this PR did nothing: it was
never a power problem, so bumping bMaxPower and disabling the brown-out detector
couldn't help (community testers steppan_5g, trill_e, wabsta, tbullas all
confirmed no change). This commit drops that experiment.

Fix: on the S3, bypass usb_new_phy() and route the DP/DM pads to the internal
OTG PHY by hand (USB_WRAP.otg_conf + RTCCNTL.usb_conf, max GPIO drive), matching
the merged CircuitPython workaround. usb_usj_mode() (bootloader entry) is made
NULL-safe because usb_init() no longer creates a phy handle on the S3.

tulip/esp32s3/usb.c is shared, so this also affects the Tulip esp32s3 (TULIP4_R11)
build; the change is guarded to ESP32-S3 and matches upstream.

Refs #952, #980.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>