STM32 CAN: Make data() return correct length slice

[adamgreig]

Currently data() always returns an 8- or 64-byte slice, even though the actual CAN frame contains less data. The remaining bytes will be zero, so the user must check the frame DLC to see how many bytes were actually sent by the remote end. With this change the slice is the same length as was transmitted, with no extra bytes.

Additionally removed the ClassicData::MAX_DATA_LEN constant because almost all instances were using a hardcoded 8 anyway, and the FdData only ever uses 64 and doesn't have an equivalent MAX_DATA_LEN. Happy to make everything use MAX_DATA_LEN instead if that's preferred.

Finally fixed a missing cfg gate on a defmt call that broke the build on G4 without defmt feature.

[Dirbaio]

seems HIL tests are failing

[adamgreig]

Oops, missed that data() was being used internally in a way that assumed it was always 8 bytes long. Fixed.

[cschuhen]

Hi guys, I saw this PR was stuck for a week. I've pushed a fix as #4137 . In it I just make raw_data public.