net: lwm2m: refactor engine to use lwm2m_message structure

Sending an lwm2m message is too difficult.  It requires pending / reply
and other structures to be configured and set by various portions of
the library.  There is also no way to know if a pending message ever
encounters a timeout.

Let's fix this by simplifying the internal LwM2M engine APIs for
handling lwm2m messages:

1. A user calls lwm2m_get_message(lwm2m_ctx) which returns the first
   available lwm2m message from an array of messages
   (total # of messages is set via CONFIG_LWM2M_ENGINE_MAX_MESSAGES).
2. Next the user sets all of the fields in the message that are
   required (type, code message id, token, etc)
3. Then the user calls lwm2m_init_message(msg).  This initializes the
   underlying zoap_packet, pending and reply structures.
4. Once initialized, the user creates their payload in msg->zpkt.
5. When the user is ready to send, the call lwm2m_send_message(msg).
6. And if for some reason an error occurs at any point, they can free
   up the entire set of structures with: lwm2m_release_message(msg).

Included in the refactoring is a timeout_cb field which can be set in
the LwM2M messages.  If a pending structure ever expires the engine
will call the timeout_cb passing in the msg structure before it's
automatically released.

Signed-off-by: Michael Scott <michael.scott@linaro.org>