Re-implement readRect to avoid RX FIFO overflow

[mudaltsov]

This fixes #29

The implementation of readRect had an issue where it would not correctly read
the last pixel from the screen. This was caused by an overflow in the RX FIFO
while waiting for the EOQ flag in the last transmitted byte. The loop only
counted transmitted bytes, so it would fill up the TX FIFO and not wait until
the transfer actually happened. Waiting for EOQ competed the queued transfers
and overflowed the RX FIFO since nothing was reading from it.

The new implementation explicitly tracks the number of RX and TX bytes to
ensure that all bytes are correctly transmitted and received. A few ideas
have been borrowed from KurtE/ILI9341_t3n as well:

• Each byte is received independently and stored in a 3-byte array, which is
  converted and written to the buffer once all 3 colors have been received.
• The TX FIFO is kept full throughout the function to avoid the delay between
  the dummy byte and the first pixel byte. Instead of using EOQ for the dummy
  byte, the results from queued commands and the dummy byte are skipped.