Instead of allocating a brand new buffer for each error string we want
to store, we can use a per-thread buffer to store the error string and
re-use the underlying storage. We already use the buffer to format the
string, so this mostly makes that more direct.

This is what it's meant all along, but now we actually have multiple
implementations, it's clearer to use the name of the library.

The old `allocfmt` is of no use to callers, as they are not able to free
the returned buffer. Export a new API that returns a static string that
doesn't need to be freed.

Bring together all of the OpenSSL initialization to
git_threads_init() so it's together and doesn't need locks.

Moving it here also gives us libssh2 thread safety (when built against
openssl).

OpenSSL's tests init everything in the main thread, so let's do that.

The OpenSSL init functions are not reentrant, which means that running
multiple fetches in parallel can cause us to crash.

Use a mutex to init OpenSSL, and since we're adding this extra checks,
init it only once.

Increasingly there are a number of components that want to do some
cleanup at global shutdown time (at least if there are not going
to be memory leaks).  This creates a very simple system of shutdown
hooks that will be invoked by git_threads_shutdown.  Right now, the
maximum number of hooks is hardcoded, but since adding a hook is
not a public API, it should be fine and I thought it was better to
start off with really simple code.