1. Replace spi.c with spi.asm
   Maximise performance as much as possible by structuring the code so
   that:
   a) The SPI transmission is initiated as quickly as possible
   b) When transferring multiple bytes, minimising the time between
      the the first byte being sent/received and the next transmission
      being initiated.
   c) Do as much as possible whilst the SPI byte is being transmitted
   Introduces spi_read() and spi_write() functions to reduce overhead
   of transmitting multiple bytes.

2. Replace sd.c with sd.asm
   Maximimise performance by:
   a) Using spi_read() and spi_write() where possible.
   b) Create static buffers for SD command sequences that are fixed
   c) Make the byte swapping of the sector address as efficient as
      possible
   d) Avoid martialing parameters off and on the stack by using non-C
      calling conventions and referencing the parameters on the
      calling function's stack
   Uses timer functions and macros instead of counted loops. This also
   makes the code to do things until a timeout occurs simpler and
   quicker. We also use compile-time convertion of milliseconds to
   number of timer counts to avoid unnecessary arithmetic.