Consider switching to embedded_hal::blocking::spi traits

[Rahix]

Disclaimer: I'm not actively using this crate right now. I was made aware of this potential problem by @Maldus512 (in Rahix/shared-bus#18) and wanted to start a discussion about it here.

Currently, the SdMmcSpi type is based on the FullDuplex trait from embedded-hal. The contract for FullDuplex has requirements on the order of calls to its methods which makes it impossible to allow sound bus sharing in the way shared-bus works. Quoting my comment in Rahix/shared-bus#18 with the reason why:

[...] the FullDuplex trait unfortunately has an API that cannot be safely shared in the way shared-bus solves sharing. The reason is that there is an implicit dependency between calls to send() and read() which could be broken if multiple bus users interweave these calls. For example:

    User 1           User 2
      |                |
    send()             |
      |                |
      |              send() - - # Now received byte from first send is discarded (or worse)
      |                |
    read()  - - - - - - - - - - # Reads data from second write
                       |
                     read() - - # HAL-API violation!!

To allow bus sharing with SdMmcSpi, I would suggest switching embedded-sdmmc to use the traits from embedded_hal::blocking::spi instead as they do not suffer this problem and can be used for bus sharing more safely (though still not concurrently across multiple execution contexts¹).

If I am not mistaken, the only place where this change would need to happen would be this method:

embedded-sdmmc-rs/src/sdmmc.rs

Lines 417 to 422 in 6071df4

	/// Send one byte and receive one byte.
	fn transfer(&self, out: u8) -> Result<u8, Error> {
	let mut spi = self.spi.borrow_mut();
	block!(spi.send(out)).map_err(|_e| Error::Transport)?;
	block!(spi.read()).map_err(|_e| Error::Transport)
	}

By using embedded_hal::blocking::spi::Transfer instead, I think it could just look like this:

    /// Send one byte and receive one byte.
    fn transfer(&self, out: u8) -> Result<u8, Error> {
        let mut spi = self.spi.borrow_mut();
        spi.transfer(&mut [out])
            .map(|b| *b[0])
            .map_err(|_e| Error::Transport)
    }

(Of course, by doing a larger change, this could be done more efficiently).