AP_HAL_ChibiOS: flush and invalidate dshot dma buffer correctly

[andyp1per]

Main problem was our invalidation was too short a length. Tested successfully on a 5"

[tpwrules]

I still think the fix should just be to the buffer length, the current strategy is correct aside from that. I understand the original issue was just missing a factor of sizeof(dmar_uint_t).

Changing the cache management strategy just wastes execution time and doesn't address any particular issue. I know this is a performance-sensitive part of the code.

[tpwrules]

For what it's worth I still would like to see just the size changes. I don't think changing the cache maintenance calls is required or necessarily a smart idea. But I don't think it's wrong.