I want to use test_olm.sh on jenkins, which lacks jq. Hence, add support to
olm.main to get id and one-time keys individually.

think this got lost when mark added separate release/debug builds

Some of the crypto libs rely on UINT64_C, which in glibc 2.17 and earlier was
not defined for C++ code (see
https://sourceware.org/bugzilla/show_bug.cgi?id=15366).

Replace the init-static-var dance with some preprocessor macros

We never delete a cipher, and the destruct op is empty, so it's a bit pointless

Give a load of internal symbols "_olm_" prefixes. This better delineates the
public and private interfaces in the module, and helps avoid internal symbols
leaking out and possibly being abused.

Should have been in the previous commit :/

These seem to be the wrong way around.

So that we can build everything together.

Split the python wrappers into separate modules

We want to stop aes_* and sha_* functions leaking out of our .so, as well as
internal _olm_* symbols.

This also means we need to link the unit tests against the objects. Possibly we
should distinguish between unit tests and integration tests.

Now that we have C and C++, we need to split the compile and link steps
(because we need different flags for the C and C++ files), so this is
easier with a Makefile.

Replace the python scripts for building the shared lib and tests with a
Makefile, which makes it easier to handle a mix of C and C++.

We don't want anything which does dynamic memory allocation in the library.

* We only use 8-byte MACs
* The version byte is \x03, not \x01

Remove the (now non-functional) declarations of olm_set_log_level in the C and
js wrappers.