Async SPI support is not straightforward

[Kernald]

Currently, async support for SPI (and implementation of the embedded-hal-async traits) is dependent on CONFIG_SPI_MASTER_ISR_IN_IRAM being disabled:

esp-idf-hal/src/spi.rs

Lines 570 to 571 in 4f44787

	#[cfg(not(esp_idf_spi_master_isr_in_iram))]
	pub async fn read_async(&mut self, words: &mut [u8]) -> Result<(), EspError> {

There are two things:

• The crate documentation doesn't make this obvious at all. One has to dig into the source to figure out why read_async and co aren't defined and figure out that it's a IDF setting.
• Async SPI support should still be provided with the interrupt handler in IRAM.

The first point is pretty straightforward to address, but addressing the second would make that irrelevant - not sure how much effort the second point actually is.

[ivmarkov]

To make the driver's async machinery operable even when the C driver's ISR processing code is in IRAM, we need to make sure, that this function is placed in IRAM as well, because spi_notify would actually be called from within the SPI C driver ISR, so it should comply to the same restrictions - i.e. the closure of all code called from an ISR handler that pretends to be "in IRAM" needs to be in IRAM too.

[ivmarkov]

The above can be done with a simple rustc #[linker_section = "something-something-iram"]. We can also conditionalize this attribute with cfg_attr to only do that if esp_idf_spi_master_isr_in_iram holds.