Optimize Internal CAN Peripheral Abstraction with DMA/etc

[dchansen06]

See https://discord.com/channels/756738476887638107/1457989912069537906/1481571048938799124

The current CAN Peripheral Abstraction copies data twice, but while the second copy is required to get the data out of the ISR and into the application, the first is to copy from the internal buffer to the external buffer which is something that DMA can do significantly faster.

It may be that the only substantial optimization is to use DMA, the etc in the title is just to be vague.

[VihanJ]

DMA will not work, must use the CPU to copy CAN messages from the message RAM.

Profiling Results:
https://discord.com/channels/756738476887638107/1457989912069537906/1489188573012824165

Calculating available deadline:
https://discord.com/channels/756738476887638107/1457989912069537906/1489406446448676966

When running CANFD bus at 5 MB/sec, must service CAN messages faster than ~250 us at 25% bus utilization, 125 us at 50% bus utilization, etc, in order to avoid overflow.

Copying 64 bytes twice (first from message ram, then within user callback) takes around 50 us. Can speed this up by allowing the user to give a location to the HAL_GetRxMessage, resulting in a single copy from message ram. Don't think this is necessary for most callbacks though, because they don't copy that much data (or its infrequent).

[VihanJ]

Scratch that, thanks to Daniel, DMA will work.

[dchansen06]

Scratch that, thanks to Daniel, DMA will work.

Specifically discussed in https://discord.com/channels/756738476887638107/1457989912069537906/1490599495317327883 and following messages