This reduces the usage of real cryptography in --cfg=fuzzing,
specifically replacing the secret->public key derivation with a
simple copy and ECDH with XOR of the public and private parts
(plus a stream of 1s to make a test pass that expected non-0
output).
It leaves secret tweak addition/multiplication as-is.
It also changes the context creation to over-allocate and store
the context flags at the end of the context buffer, allowing us
to easily test context flags in each function.
While it would be nice to have something fancier (eg XOR-based),
its not immediately obvious how to accomplish this, and better to
fix the issues I have than spend too much time on it.
Fixes#271.
This partially reverts b811ec133a
We can now run unit tests with the fuzz feature on, and they'll pass,
which is some assurance that fuzzing with the feature on won't lead to
spurious failures due to the fuzz harness inadequately simulating message
signing.
It's super dangerous to use Cargo features for this, since they can be set
accidentally (or maliciously by any crate in a user's entire dep tree). Instead
we can just require users set `RUSTFLAGS` appropriately, which we can easily
do in our fuzzing scripts.
This feature was not useful for Cargo users, since Cargo does not give you
the kind of fine-grained control over C library linkage that you need. So
it was just unnecessarily confusing and would cause the build to break if
you enabled it accidentally, say, with --all-features.