spi: stm32: use STM32 DMA with STM32 MDMA to enhance DDR use

spi: stm32: use STM32 DMA with STM32 MDMA to enhance DDR use

The STM32 DMA doesn't have the ability to generate convenient burst
transfer on the DDR, ensuring the best load of the AXI & DDR.
To avoid this bad load of the AXI & DDR, STM32 MDMA can be used to transfer
data to the DDR, being triggered by STM32 DMA channel transfer
completion.
An SRAM buffer is used between DMA and MDMA. So the MDMA always does
MEM_TO_MEM transfers (from/to SRAM to/from DDR), and the DMA uses SRAM
instead of DDR with DEV_TO_MEM transfers.
SPI RX DMA (DEV_TO_MEM) becomes:
SPI RX FIFO ==DMA==> SRAM ==MDMA==> DDR

In RX (DEV_TO_MEM), EOT interrupt is used to pause the DMA channel (which
will raise a transfer complete) to trigger the MDMA to flush the SRAM (when
transfer length is not aligned on SRAM period).
TX remains on the former implementation.

Signed-off-by: Clément Le Goffic <clement.legoffic@foss.st.com>
Link: https://patch.msgid.link/20250616-spi-upstream-v1-4-7e8593f3f75d@foss.st.com
Signed-off-by: Mark Brown <broonie@kernel.org>