feat(drivers): port DP3246 shift driver initialization

[stuartparmenter]

Summary

• Replaces the DP3246 stub (which logged "DP3246 initialization not yet implemented" and did nothing) with the actual bit-bang init sequence: pre-clear sweep → REG1 (max output current, no OE widening) → REG2 (max blanking potential, single-edge transfer, power-saving features off) → trailing blank → OE enable. New file: components/hub75/src/drivers/dp3246.cpp.
• DP3246 needs rising-edge clock. HUB75_CLK_PHASE_INVERTED now also default y if HUB75_DRIVER_DP3246 (mirrors the MBI5124 default). For users who construct Hub75Config directly in code, the dispatch in DriverInit::initialize logs a warning when clk_phase_inverted is left false.
• Internal helper shift_with_latch(pins, pixels_per_row, pattern, lat_cycles) collapses the four near-identical "shift N pixels, raise LAT for the last K, drop LAT" sweeps in dp3246_init into a single parameterized call. Latch widths are named (LAT_CYCLES_BLANK = 3, LAT_CYCLES_REG1 = 11, LAT_CYCLES_REG2 = 12).

Known limitation (called out in code, log, and Kconfig help)

DP3246 also expects LAT held high for 3 clock cycles per row at runtime, not just during init. The DMA backends (gdma_dma, i2s_dma, parlio_dma) currently emit a 1-cycle LAT pulse on the last word of each row, so panels may still show artifacts after this init succeeds. Widening the runtime LAT pulse is intentionally left to a follow-up PR — that change touches all three DMA backends and adds a new Hub75Config::latch_pulse_width field, which is too much for an init-only port.

Test plan

• Compile-only: build simple_colors example with CONFIG_HUB75_DRIVER_DP3246=y for ESP32, ESP32-S2, ESP32-S3, ESP32-P4, ESP32-C6 — confirm no missing symbol / link errors and that HUB75_CLK_PHASE_INVERTED flips on automatically.
• Confirm FM6126A / FM6124 / GENERIC paths are unchanged (DP3246 is a separate case branch; only addition is the conditional clk_phase_inverted warning).
• On hardware (DP3246 panel, when available): verify the init sequence runs without watchdog trip and that the warning about runtime LAT widening fires once at startup. End-to-end correct rendering will require the follow-up DMA-side LAT widening.