blk: Make the i.MX8 mmc driver more robust to errors

[midnightveil]

• The in-tree code relied on a lot of assertions and would crash in
  the presence of any sort of errors, e.g. card removal.

• This new implementation handles command errors uniformly in a new
  handle_interrupt_status() function, and returns varying error types
  up the call stack and to the block queues.

Further/more detailed changes are:

• The INT_STATUS register is now only read once on an interrupt, to
  prevent subtle race conditions.

• Soft failures from:

  • Card not expecting an APP_CMD following CMD55
  • Incompatible voltage ranges, or unsupported card types
  • Incorrect card check commands from CMD8

• sDDF Block queues are only ever initialised once

• Indefinite busy-wait loops on Command Inhibit fields are replaced
  with an error return, as they never happen in usual operation.
  Similarly, busy wait on DAT[0] following R1b commands are handled
  by the IP, so it never happens.

• On failed initialisation, reset (most) of the driver state, so we can
  try again in future. This is prep work for hotplugging support.

• Add a stub card_detected() function which currently always returns
  true, but if implemented will correctly send "Gone" errors responses.

• Add BLK_RESP_DEV_GONE error to the block queue
  This handles the case where the block device has been unplugged
  or is otherwise unavailable.

• Add -MD to the example CFLAGS to track header-file dependencies

Split out the error-handling code from #180 to make my life easier.