fix(power): arm deep-sleep hold so per-pin gpio_hold_en actually holds

[DatanoiseTV]

doDeepSleep called gpio_hold_en for BUTTON_PIN and LORA_CS, but the
ESP-IDF API contract is that gpio_hold_en alone only covers light
sleep — deep sleep additionally requires gpio_deep_sleep_hold_en
once before esp_deep_sleep_start to arm the holds across the deep-
sleep boundary. Without it the pins float in DSLP, which on some
boards leaks current through LORA_CS and on others triggers
spurious wake from a noisy button line.

Split out from #10425 — single-concern PR.

Build verification

pio run -e t-deck-tft succeeds, no new warnings.

Attestations

• I have tested that my proposed changes behave as described — review/static-analysis only, not on-air.
• On-hardware testing requested from community: build-verified t-deck-tft only.

[Copilot]

Pull request overview

This PR fixes ESP32 deep-sleep GPIO retention by explicitly arming the global deep-sleep hold mechanism so existing per-pin gpio_hold_en() calls (e.g., BUTTON_PIN, LORA_CS) actually persist across the deep-sleep boundary. This prevents pins from floating during deep sleep, which can cause excess current draw and/or spurious activity on some boards.

Changes:

• Enable ESP32 deep-sleep GPIO holds in doDeepSleep() via gpio_deep_sleep_hold_en() immediately before entering deep sleep.