Bdot

The release fixes a bug with small carrays.

Running bdot on a carray whose length was smaller than it's chunklen caused a segfault :hankey:. This release fixes this bug.

This release should have enough features to be useful, and few enough issues to be worth it.

Enjoy!

Features
- speed on par with `numpy.dot`
- input and output can be bigger than memory (via compression)
- input and output directly (and automatically) to and from disk

Details
- `dot` for `bcolz.carray`
- `matrix . vector`
- `matrix . matrix`
- `out` parameter
- write results to `bcolz.carray`
- write results directly to disk

Issues
- precision on `float32` could be better
- balancing compression ratio and compute performance is still very manual

Usable but likely to have issues in some important areas, including:
- space and time characteristics based on:
- `chunklen`
- `clevel`
- precision on `float32`