TLS compliance: record fragmentation

[bugadani]

This was mentioned in #75 (comment) but that issue has since been closed.

[lulf]

Does it need to for 99% of embedded use cases?

[bugadani]

No idea, but on one hand, it might help with MaxFragmentLength extension to tune things, and on the other, I just want this fact visible.

Edit: my guess here is that it's probably not necessary, but also that we might implement something for it under the alloc flag. There's no reason not to be compliant if there is a way to do so :) Definitely not something I'll spend time on any time soon, though.

[newAM]

I have started working on this in #160.

The most difficult part of adding record fragmentation is testing. I have found TLS implementations that will frequently fragment, but getting them to fragment reliably for testing purposes can be difficult. For testing this feature I wrote another TLS stack newAM/tls-tester to generate heavily fragmented records. In the future I hope to expand tls-tester to test overall TLS compliance and implement all requirements for TLS 1.3 compliance in embedded-tls, but record fragmentation is a large enough goal for now.

Implementing fragmentation will require a lot of code changes. I am implementing this for a hobby project that has no deadlines, and I am likely to take several months-long breaks while working on it, but at least there is a way to test fragmentation now if someone else was to pick up this work.