Summary

This PR gives a speed boost to simple SPI.transfer(data) and SPI.transfer(buffer, count) operations
in which the user controls the chip-select by directly using only the necessary LL-HAL functions and not going thru the CS controlling functions and the spi_transfer() function.

With no DMA support in the SPI class the sending of a buffer takes a long time and with this modification
it takes at least less time.

Measured with STM32C031C6, SPI clock 8MHz and a buffer of 231 bytes:
digitalWrite(CS, LOW);
SPI.transfer(&buffer, 231);
digitalWrite(CS, HIGH);

original code: 587,26µs
this pr: 474.92µs