Gbatchy

- Add chain() which can be used as a low-overhead way to build more efficient sequences without spawning a greenlet.
- Small performances optimizations for redis (__slots__)
- Upgrade to gevent 1.1+, while keeping the gbatchy exc_info behavior (you get the whole traceback once).

- Small performance optimization for redis.
- Small performance optimization for transform() (& by proxy batched functions)

- Fix a context propagation via add_auto_wrapper when using batch functions. Also makes batch functions slightly more efficient.

- Added a may_block context manager to be able to use gevent primitives between batch greenlets. For an example, see the iwait & wait implementations.
- Add a version of iwait & wait that work in a batch context.
- Fix *_unordered - this was previously using gevent.iwait.
- Add a Pool implementation that mirrors gevent.pool.Pool, but works with batch greenlets. Unfortunately it was not as simple as greenlet_class=BatchGreenlet.

- Fix redis *scan functions.

Features:
- A couple of useful utils: `immediate`, `immediate_exception`, and `transform`.
- `pmap`/`pfilter` functions now take `**kwargs` that get passed through.
- New function `spawn_proxy`: same as `spawn`, but returns a proxy-type that evaluates to spawn().get()
- Add `as_future=True` to batched/class_batched functions. Since the batched system
uses AsyncResult-like objects under the hood anyway, this lets you take advantage of
that. You can save a bit of spawn() calls.
- Add max_size to specify the maximum number of calls to coalesce.
TODO: Add a way to specify this as e.g. sum(map(len, args_list))

Bugfixes:
- BatchAsyncResult.wait was broken if the value was not available at the time of the call.
- BatchAsyncResult.get has support for the timeout arg.
- BatchAsyncResult.get_nowait exists (alias for .get(block=False))
- BatchAsyncResult.set_exc_info now properly sets .exception (in case .get() is called multiple times).
- BatchAsyncResult.successful no longer raises.
- Circular dependency between context & scheduler is now gone.