Great refactoring (and co-maintainer request)

[jerry73204]

The PR is almost a total rewriting of the crate. The highlight is that it adopts AlexeyAB's darknet instead of pjreddie's, due to the fact that the original author no longer maintains it. Several changes are made and issues are fixed in this patch.

• Combined with newer darknet-sys 0.2.0. The PR requires new patches on darknet-sys to adopt AlexeyAB's darknet.
• Better integration with image crate. Now the generic ImageBuffer and the crate's Image can convert to each other by From and TryFrom traits.
• Fix memory leakage issue. It's mostly due to unmatched {,de}allocations. For example, Drop::drop on a calloc() from C.
• The new Network::layers returns a iterable collection of layers for inspection.
• New train_detector() method to train a model.
• The new Error type implements Fail trait for fluent integration with failure crate.
• More convenient methods are added to existing types Detection::best_class, Network::input_hape, etc.
• The original YOLOv3 tiny example is modified in the way that it automatically downloads the weights. A new example run_inference is added to test existing configurations and weights.
• Better documentation: More doc comments and a verbose README.

The version is bumped to v0.2.0 due to breaking changes. I would be glad if these changes can be published to crates.io eventually. If possible, please promote me as the co-maintainer. Thanks.

[jerry73204]

Thanks!

[jerry73204]

I think it's time to push it to crates.io to make it publicly available.