stm32/boards/OPENMVPT: Fix white illuminator default state and Lepton reset

[kwagyeman]

Two independent OPENMVPT board fixes, both found while debugging an LTDC
issue on a bench board and both verified on hardware.

Stop turning the white illuminator on at boot

MICROPY_HW_LED4 mapped the white LED to PG3, but that pin is not wired
like the other three. Red, green and blue drive an HSMF-C116 whose common
pin ties to 3V3 through 220R, so they are active low, which is what the
board's MICROPY_HW_LED_ON()/OFF() macros assume. PG3 instead gates a
PMZ390UN N-channel MOSFET with a 100K pulldown on its gate, so the white
LED is active high.

So MICROPY_HW_LED_OFF() drove PG3 high and switched the MOSFET on:
led_init() lit the illuminator on every boot and left it on. It is not a
status LED, it is a CLP6B-WKW rated 3 x 50mA drawing 100mA through R120 off
the 5V rail.

The polarity macros are board-wide and correct for the RGB LED, and the
stm32 port has no per-LED polarity hook, so PG3 is left out of the LED
table. Nothing then configures the pin, it stays high impedance out of
reset, and R142 holds the gate low.

This also removes an inverted API: pyb.LED(4).on() turned the illuminator
off and .off() turned it on. No example uses LED(4) on this board.
Pin("LED_WHITE") remains available for deliberate control.

Verified on hardware: PG3 comes up mode=analog, ODR=0.

Actually reset the Lepton at boot

board_early_init() drove LEPTON_RSTN straight high and never pulsed it
low, and released it before starting MCLK. Both are wrong for this part: it
wants its master clock present when RESET_L is released, and it never
receives a reset at all.

A cold power-up POR hides this. Once the Lepton is confused it becomes
unrecoverable: it still ACKs its I2C address so the probe detects it, but
its status register never reads LEPTON_I2C_STATUS_BOOT, so
lepton_config() times out and omv_csi_init() returns
OMV_CSI_ERROR_CSI_INIT_FAILED. main() treats anything other than
OMV_CSI_ERROR_ISC_UNDETECTED as fatal, and this happens before USB is
brought up, so the board stops enumerating entirely - no CDC, no REPL, no
way in short of SWD - and every subsequent MCU reset runs this same code
and leaves the part exactly as it was.

Observed on a bench board that stayed in this state across MCU resets, an
nRST pin reset, toggling the shared CSI power/reset rail, and a physical
power cycle. Pulsing RSTN low by hand brought it straight back, which is
what led to the fix.

Hold RSTN low, release powerdown, start MCLK, spin past the 5000 clock
periods the part requires, then release RSTN. SystemClock_Config() has
not run this early, hence the spin rather than mp_hal_delay_ms().

Verified on hardware: Lepton enumerates at 160x120, thermal frames capture,
and the full thermal overlay demo path runs at 17.9fps.