Bluetooth: Mesh: Refactor CCM

Unifies the Mesh CCM implementation parts for encryption and decryption
into a crypt and an auth step, reducing stack usage and code size.

This change also brings several performance improvements, most notably
reducing copying of the nonce and unrolling the 16 byte XOR operations.

Performance for the Mesh worst case of a 382 byte payload with 16 bytes
of additional data (full transport encrypt with virtual address) goes
from an average of 889us to 780us on nRF52840 with default optimization
flags.

Signed-off-by: Trond Einar Snekvik <Trond.Einar.Snekvik@nordicsemi.no>