crypto/ccm: bugs in the implementation of CCM mode

[cladmi]

This issue summarize all the issues I have found when trying to add tests for #6706 for sys/crypto/modes/ccm.c

Issues shown in the PR:

• Upper bound for auth_data_len should be 0xFEFF. https://tools.ietf.org/html/rfc3610 2.2 page
• Wrong check for input plaintext message length against len encoding.
• buffer overflow if auth_data_len is larger than 26

Problems found when adding tests:

• In tests 'len_encoding' and 'mac_length' are hardwritten.
• Tests do not support input_len >= 256
• ccm code will break if length_encoding >= 4 (even with PR fix because of overflow).
• Code is not saving enough bytes when saving plaintext_length (off by one limit)
• CCM is broken with input_len >= 256 (so broken even when fixing previous point)

• Refactor test to allow running other tests (tests-crypto-modes-ccm: Refactoring to support more evolved tests #8109)
• Fix and add test for input_len check (crypto/ccm: fix input_len checking with length_encoding #8112)
• Fix and add test for auth_data_len upper bound (crypto/ccm: fix auth_data_len check #8114)
• Fix and add test for auth_data_len > 26
• Fix and test for input_len >= 256
• Test working on 8bit/16bit platforms…
• Add all RFC/NIST tests tests/sys_crypto: add RFC3610 and NIST SP 800-38C tests #12038
  • Make the api take const inputs crypto/modes/ccm: update api to const input buffers #10376
  • split tests-crypto out of unittests (part of tests/sys_crypto: add RFC3610 and NIST SP 800-38C tests #12038)

This is to track and fix all what was done in #7968.
The PRs will use and replace #6706

[fjmolinas]

@aabadie can we check the 8bit/16bit point since #12645 is in?