There were a few things wrong with the SD card write implementation:
1. The protocol change regarding the interpretation of offsets
introduced with SDHC cards was not taken into account in the write
path, unlike in the read path.
2. All SPI writes involved in the process were actually issued as reads
due to the use of the SPI_RW() function, the implementation of which
seems to have gone through some churn. Likely just an instance of
bit-rot.
