feat(server): check box certificates

[Oaradla]

Builds on #534 (its commit is included here until that one is merged).

Until now the box port accepted any client certificate whose common name looked like a box. Now:

• Certificates issued by a TeddyCloud CA (TB1, TB2) are trusted. Such a box is pinned too, so no self-made
  certificate can take its place later.
• Original Boxine certificates have no checkable chain (the box only sends its own certificate, issued by "Boxine
  Factory SubCA nn", which is not available). So they are checked against the real client certificate: if the
  original client.der of the box is uploaded, the box has to present exactly that certificate (also when another
  one was pinned before); without it, the first certificate a known box presents is pinned in toniebox.certPin,
  also with core.allowNewBox off, so nobody is locked out by the update. allowNewBox still decides about unknown
  boxes.
• The existing core.boxCertAuth (default on) refuses boxes whose certificate does not match; off, they are only
  logged.
• Result per box in internal.boxCertStatus (trusted / pinned / mismatch) and internal.boxCertSha256; a pin is
  reset with /api/settings/reset/toniebox.certPin?overlay=<box>.
• Same check for the MQTT server: a refused box gets CONNACK "not authorized". An unknown box id only counts as a
  box there while allowNewBox is on (before, the MQTT server created an overlay for any box id).

Issuer check and box id from the certificate subject in box_cert.c (C tests), the check in box_cert_check.c;
the TLS context keeps the SHA-256 fingerprint of the box certificate.

Tests: test_box_cert_check.py (trusted, pinned on first contact, mismatch refused, only logged without
boxCertAuth, pin reset, known box pinned with allowNewBox off, uploaded client.der wins over an earlier pin, a box
with a TeddyCloud certificate refuses a self-made one).

Device-tested with the original firmware of a CC3200 box (3.1.0 BF2, 3.3.0, 3.4.0: pinned on first contact,
refused with a wrong pin) and an ESP32 box (v5.237.0: pinned, the same certificate as the client.der extracted
from its flash).

Refs #85

[henryk86]

So in short just for my understanding.

A box is still identified by its mac, but now it's additionally checked that the boxes certificate matches the pinned one (first contact original tonies cert or teddycloud cert). So if some bad guy knows the Mac and creates a fake cert it will reject?

[Oaradla]

Yes, exactly. The MAC from the certificate's common name still decides which box it is. On top of that, the certificate itself now has to be the right one:

• Certificate issued by TeddyCloud (e.g. TB2): it is verified against the TeddyCloud CA, so it can't be faked without the CA key. The box is pinned to it as well, so a self-made certificate for that MAC is refused later.
• Original Boxine certificate: there is no CA to check against (the "Boxine Factory SubCA" isn't available), so it is compared with the real one. That is either the client.der you uploaded or extracted for that box, or, without one, the certificate the box presented on its first
  contact after the update (pinned in toniebox.certPin).

So someone who knows the MAC and builds their own certificate gets a different fingerprint, which counts as a mismatch: 401 on the box port, "not authorized" on MQTT (with core.boxCertAuth, which is on by default). Copying the real certificate doesn't help either: the TLS handshake
needs the box's private key.

One gap remains: if no client.der is uploaded and an attacker connects with that MAC before the real box does, the attacker's certificate gets pinned. The real box is then refused and shows up as "mismatch" in the log and the web UI. Resetting the pin
(/api/settings/reset/toniebox.certPin?overlay=) or uploading the original client.der fixes that. Uploading the client.der closes this window from the start.

[SciLor]

#533, #534, #535 and #537 are merged. Could you rebase this onto develop? The #534 commit can be dropped, and #535 touches the same three lines in Makefile, tests/c/test_runner.c and tests/c/tests.h. After that I'll merge it.

[Oaradla]

Rebased onto develop: one commit now, the #534 commit is dropped. Makefile, tests/c/test_runner.c and tests/c/tests.h keep both the esp32_port entries from #535 and the box_cert ones. The rest of the diff is unchanged. Build and tests pass (C tests, box_cert_check, box_cert_upload, box_sim).

[SciLor]

Could you please rebase again / merge as needed?

[Oaradla]

Rebased onto develop again (after #545).