Pyrodigal

Changed
- Make `Pyrodigal.train` return a reference to the newly created `TrainingInfo` for inspection if needed.
- Reimplement `add_nodes` and `add_genes` to use a growable array instead of counting and pre-allocating the C arrays.

Fixed
- Inconsistent handling of unknown nucleotides in input sequences and gene translations.

Added
- Additional `Gene` properties to access the score

Changed
- Use more efficient `PyUnicode` macros when reading or creating a string containing a nucleotide or a protein sequence.
- Release the GIL when creating a bitmap for an `str` given as input to `Pyrodigal.find_genes`.
- Release the GIL when creating the protein sequence returned by `Gene.translate`.

Fixed
- `Pyrodigal.find_genes` and `Gene.translate` not behaving like Prodigal when handling sequences with unknown nucleotides.

Added
- `pyrodigal.TrainingInfo` class exposing variables obtained during training as an attribute to `Pyrodigal`, `Gene` and `Genes` instance.
- Support for passing objects implementing the buffer protocol to `Pyrodigal.find_genes` and `Pyrodigal.train` instead of requiring `str` sequences.

Fixed
- Potential data race on training info in case a `Gene.translate` with a non-default translation table was being translated at the same time as a `Pyrodigal.find_genes` call.
- Spurious handling of Unicode strings causing potential issues on platform using a different base encoding.

Fixed
- `Pyrodigal.find_genes` segfaulting on some sequences when called in `single` mode ([2](https://github.com/althonos/pyrodigal/issues/2)).
- `MemoryError` potentially not being properly raised on allocation issues for sequence bitmaps.

Changed
- Tests are now in the `pyrodigal.tests` module and can be run after a site install.

Fixed
- `Pyrodigal.find_genes` stalling on sequences shorter than 3 nucleotides.

Fixed
- Compilation of OSX and Windows wheels.