Investigate C-INTERMEDIATE for all stabilized drivers

[bugadani]

cc #2784

[bjoernQ]

To avoid breaking changes in future:

To resolve this I guess for the non-error case we need to carefully look at every function returning ! or Result<(),E> (or where Result::T is a simple type)

For the error case we could consider that every error variant contains a payload, e.g. MaxDmaTransferSizeExceeded(MaxDmaTransferSizeExceededInfo), - initially that's just an empty struct (non-exhaustive) - if we ever consider to provide more information we could extend that struct without a breaking change

However, I'm not sure that is worth it

[MabezDev]

So I believe this only applies to I2C & SPI, as they are the two drivers with the concept of a transaction.

I think to take @bjoernQ's ideas a bit further we should:

We should change the return type of spi/i2c transactions to be Result<(), TransactionError>, where TransactionError looks like:

pub struct TransactionError {
   inner: Error,
}

Initially we'd only expose the error inside TransactionError, but it would create a path for us to give some more context to the error when multiple operations may have occurred. For example adding sucessful_operations: usize and a method to access it.

The thing is, embedded-hal only allows us to have on error type for the whole driver, so this will never be available with the eh api. So here we have the option of marking the inherent transaction methods on the drivers as unstable, and we can punt this to post 1.0. Thoughts?