- Fix the memory corruption caused by the wrong MPU region configuration, which
enabled caching of the shared memory region, without any cache maintenance in
libmetal.
- Fix the MPU region's TEX level, which was setting the shared memory region as
a device and/or strongly-ordered memory, which caused usage faults on unaligned
accesses.
- Fix a deadlock that occurs if the main thread (sketch) makes a remote call
and holds the virtqueue's mutex, and then gets preempted by the event thread,
which in turn tries to acquire the same mutex. Note that libmetal's mutex
implementation doesn't yield the CPU, it can be changed to do so, but for now
the queues are protected with an OS mutex.
- Fix RPCLIB msgpack unpacker memory overflow if RPMSG endpoint buffer > 512.
- Optimize RPCLIB msgpack unpacker usage by allocating once and extending on
demand.
- Remove extra request/response threads, locking and callbacks, that were
causing all sorts of concurrency issues, and were detrimental to performance.
- Remove the redundant copy of messages to ring buffer, to boost the performance.
- Remove many smaller dynamic allocation to improve the performance.
- Use a single endpoint for message packs requests/responses ("RPC"), and
another one dedicated for raw (non-packed) data.
- General refactoring and cleanups, removal of dead code and variables,
debugging code, etc...
Signed-off-by: iabdalkader <i.abdalkader@gmail.com>

- Fix serial pass-through example.
- Add MD5 checksum example.
- Add RAW-channel ping-pong example.
Signed-off-by: iabdalkader <i.abdalkader@gmail.com>