JNI rebased

[greenaddress]

This is another continuation of #64 and #248 implementing JNI library for secp256k1, rebased to master.

Also it includes a fix for a bug we noticed while including it in GreenBits - buffers were overflowing if two different functions were called consecutively with the second one needing more buffer space.

[ghost]

Thanks for pulling this forward and testing. It looks like Travis is still having a few problems in picking up the jni object, but I'm going to close my original PR in favor of this, as its rebased and tests are passing (running make check-java) locally at least

[greenaddress]

@faizkhan00 thank you and @theuni (and anyone else involved I missed) for the work done thus far.

tests are passing locally but there seems to be a race condition in make where by the shared library is not fully linked by the time the test starts (but strangely only in travis, we can't reproduce locally) - looks like we need to fix the dependency

we tried to fix that and failed (we are not too strong on makefiles) so we asked @theuni and he suggested he may have a look at this at some point but anyone with any input is welcome

[theuni]

@greenaddress Sorry for not getting to this at the meetup as promised, I kept getting pulled away.

Anyway, that's probably for the best. This turned out to be non-trivial to track down, and not build-system related at all.

See the top few commits here for the fix: https://github.com/theuni/secp256k1/commits/jni-test

tl;dr: JNIEXPORT on some Java versions (the one Travis uses by default, at least) is busted wrt symbol visibility.

[greenaddress]

@theuni no worries! Cheers for finding the actual source of the problem!

[sipa]

Concept ACK, but can you rebase and squash the fixups?

[sipa]

This function can legitimately return 0, if the passed signature's R coordinate does not correspond to a point on the curve (around 50% chance for arbitrary byte arrays).

[greenaddress]

secp256k1_ecdsa_recover was not implemented in org_bitcoin_NativeSecp256k1.c, hence the TODO and commented out code.

I've now removed the unimplemented and commented parts and cleaned it up in general. I think it's reasonable to keep these two commits for future reference, in case someone wants to look up the incomplete implementations in git history.