Abstract the orion_nand_fast_block_write() routine into a separate
routine -- arm_nandwrite() -- so that other ARM cores can reuse it.

Have davinci_nand do so.  This faster than byte-at-a-time ops by a
factor of three (!), even given the slowish interactions to support
hardware ECC (1-bit flavor in that test) each 512 bytes; those could
be read more efficiently by on-chip code.

NOTE that until there's a generic "ARM algorithm" structure, this
can't work on newer ARMv6 (like ARM1136) or ARMv7A (like Cortex-A8)
cores, though the downloaded code itself would work just fine there.

git-svn-id: svn://svn.berlios.de/openocd/trunk@2663 b42882b7-edfa-0310-969c-e2dbd0fdcd60