Having the keys on the stack increases the stack space consumption by
quite a bit, and this in turn results in the sanitizer tests failing
for Rainbow.

Moving the keys to the heap in the test seems like a harmless change.

* Add state destroy to SHA2 API

* Include optimized SPHINCS+ implementations

I've generated new implementations from the sphincsplus repository.

* Don't destroy sha256ctx after finalize

* Attempt to shut up MSVC

* Make sure to drop errors in rmtree

* remove automatically checked property

* Allow PR-submitters to self-document.

We don't ever use these, so allow submitters of PRs to self-document.

* Add ccache to speed up builds

* Fix circleci config

* fixup! Fix circleci config

* Try again with environment variables

* Set up ccache on MacOS

Every .s file is now handled through CC

CircleCI has AVX512 support and the compiler will generate instructions that Valgrind doesn't handle.

In light of #251

clang-tidy9.0.0 added a new check: bugprone-branch-clone
(https://releases.llvm.org/9.0.0/tools/clang/tools/extra/docs/ReleaseNotes.html)
This doesn't like both branches of an if are the same.
This lead to a warning in rainbow, as where the maximum of two values (which
are always the same) is computed in a macro.
I don't always agree with this warning, but here I think it's worth to
remove the macro.

clang-tidy9.0.0 added a new check: bugprone-branch-clone
(https://releases.llvm.org/9.0.0/tools/clang/tools/extra/docs/ReleaseNotes.html)
This doesn't like both branches of an if are the same.
In this case I don't think there is any reason to do this, so I've removed it.

clang9.0.0 (https://releases.llvm.org/9.0.0/tools/clang/docs/ReleaseNotes.html)
adds a new satic analyzer: security.insecureAPI.DeprecatedOrUnsafeBufferHandling
which throws warnings if you use "unsafe" buffer handling functions which
includes memset and memcpy.
We have memset and mempy all over the place, so I think it's best to ignore this warning.
All the occurences that I looked at seemed perfectly "safe" to me.

* Copied qTESLA-p-I round2 (2019-08-19) code

* Code compiles, NIST-KAT works

* Included detached signature API

* Generated testvectors

* Fixed name in api.h

* code style

* Fixed error in Makefile

* Passing pytest

* Fixing types (uint8_t bytes and size_t indices)

* Replaced SHAKE with SHAKE128 where necessary

* Fixed bug: (signed) integer overflow

* Added qTESLA-p-III

* Code is now independent of machine endianness

* repaired Microsoft makefile

See https://circleci.com/gh/PQClean/PQClean/7991

Fix kyber-90s warning if size_t is not 32 bits

Update Kyber-90s to use aes256_ecb instead of local AES implementation

It seems wise to include _some_ form of guidance on what we know about
problems with certain (versions of) implementations.

[ci skip]

Fix and re-add Rainbow