Let the destination bus decide DMA for sprites; fix clipped images on Panel_AMOLED

[ainyan03]

Let the destination bus decide DMA for sprites

• pushSprite passed SpriteBuffer::use_dma() to the panel, deciding on the caller side whether the sprite buffer could be read by DMA. The bus already checks the source itself (SPI / parallel copy or use the CPU when the source is not DMA-capable), so pushSprite now always passes true and use_dma() is deprecated. Nothing new is read by DMA on the ESP32 family.
• The meaning of use_dma is documented on Bus / Panel / LGFXBase::pushImage: true allows the bus to keep reading the caller's buffer until the transfer completes; false means the buffer is not read after the call returns. Wait on the destination with waitDMA() before reusing the buffer while startWrite() is held. pushImage keeps false and pushImageDMA keeps true.

Panel_AMOLED: clipped images

• Rows of a horizontally clipped image were sent separately: with use_dma they were queued without the pixel-write start, and without it every row restarted the write (RAMWR), which rewinds to the window start. On a CO5300 panel driven without the frame buffer, a clipped pushSprite / pushImage drew nothing. The write is now started once and all rows go in that frame. (StopWatch normally draws through its frame buffer, so it was not affected in normal use.)

Testing

• StopWatch with the frame buffer bypassed: a sprite clipped to its left half (internal RAM sprite, flash setBuffer sprite, pushImage from flash) matches the unclipped reference; before this change the clipped cases drew nothing.
• Local build matrix of this repository's and M5Unified's CI versions (Arduino 2.x / 3.x, ESP-IDF 5.1 to 6.1, all chips) plus the native (SDL) build, with no deprecation warning from the library. CI on the fork: passed.
• The same change is proposed to LovyanGFX (the shared lgfx/v1 sources).