Clear stale open-drain pad state in QSPI, parallel bus, and PWM backlight init

[ainyan03]

Follow-up to #251 (#192). ESP-IDF v6 delegates pad state management (open-drain etc.) to the peripheral user, and an audit of the remaining init paths found three more places that relied on driver side effects that no longer exist on v6 — or never existed on any version:

1. spi::initQuad (QSPI panels) — same issue as M5Stack CoreS3 display (ILI9342C) blank on ESP-IDF v6.0.1 #192: spi_bus_initialize() on v6 no longer clears the pad open-drain flag, and Fix CoreS3 LCD not working on ESP-IDF v6 (SPI bus pins left in open-drain) #251 covered only the single-bit spi::init path. The pad restoration is factored into a shared clear_open_drain() helper and applied to SCLK/IO0-3 as well.
2. Light_PWM::init (PWM backlight) — the LEDC driver never clears the open-drain flag (on any IDF version), and v6 no longer selects the GPIO function in IO_MUX either. The pin is now explicitly normalized as a push-pull GPIO output at the start of init, latched to the "off" level first to avoid a visible glitch.
3. Bus_Parallel8/16 on ESP32-S2/S3 — neither esp_lcd_new_i80_bus (S3; its gpio_set_direction call was removed in ESP-IDF v5.4) nor the plain GPIO matrix routing (S2) configures pad direction / open-drain on the data pins. The same pinMode(output) normalization already used by Bus_EPD is applied. The WR/RD/RS control pins were already normalized via gpio_set_direction and are unaffected.

A fourth commit makes Light_PWM::init report failure: it previously ignored the LEDC API results and always returned true. It now fails early for pins that cannot be driven as outputs and propagates LEDC setup errors. (Panel_Device::init ignores the return value, so panel initialization behavior is unchanged.)

Verification

Each fix was verified A/B on hardware, with the pads deliberately set to open-drain before init (the state left behind by e.g. probing pins as inputs), checking the GPIO_PINn.pad_driver register after init:

• initQuad: M5StopWatch (CO5300 QSPI panel) + ESP-IDF v6.0.1 — without the fix the pads remain open-drain after init; with it they are restored to push-pull. No regression on v5.5.4.
• Light_PWM: M5Stack Core FIRE (backlight on GPIO32) + ESP-IDF v6.0.1 — same A/B result.
• Bus_Parallel8 (S3): scope-free rig counting bus edges with PCNT — with the data pins left open-drain, a 0x00/0xFF test pattern produces 0 edges on D0/D7 while WR strobes normally; with the fix the expected edge counts appear.
• Builds: Arduino-ESP32 2.x / 3.x (esp32, esp32s3), ESP-IDF v6.0.1 (esp32, esp32s2, esp32s3), ESP-IDF v5.5.4 (esp32s3).
• CI on the fork: Arduino / ESP-IDF build workflows green.