Pyhmmer

Added
- `plan7.ScoreData` class to store the substitution scores and maximal extensions for a long target search.
- `plan7.LongTargetsPipeline` to run searches on targets longer than 100,000 residues.
- `Alphabet` methods to check whether an `Alphabet` object is a DNA, RNA, nucleotide or protein alphabet.
- `window_length` and `window_beta` arguments to `plan7.Builder` to set the max length of nucleotide `HMM` created by builder objects.

Changed
- `pyhmmer.hmmer.nhmmer` now uses a `LongTargetsPipeline` instead of a `Pipeline` to search the target sequences.
- `pyhmmer.hmmer.nhmmer` now supports `HMM` queries in addition to `DigitalSequence` and `DigitalMSA` queries.
- `pyhmmer.hmmer.phmmer` now always assumes protein queries.
- `Z` and `domZ` attributes of `plan7.TopHits` objects is now read-only.

Fixed
- `nhmmer` now uses DNA as the default alphabet instead of amino acid alphabet like it did before ([12](https://github.com/althonos/pyhmmer/issues/12)).

Added
- Constructor arguments and properties to `plan7.Pipeline` to support bit score thresholds instead to filter top hits.
- Support for creating a `SequenceFile` and an `MSAFile` using a Python file-like object instead of only supporting filenames.
- Support for reading individual sequences from an MSA file with `SequenceFile`.
- `TextMSA.alignment` to access the actual alignment as a tuple of strings.
- Subtraction and division support for `easel.Vector` subclasses

Changed
- `plan7.Cutoffs` now support setting the bit score cutoffs, but requires both to be set or cleared at the same time.
- `easel.Vector` will always allocate some memory when created manually to avoid having a special empty case in every vector method.
- `pyhmmer.easel.AllocationError` now stores the size it failed to allocate, and the number of elements when allocating an array.

Fixed
- `TextSequence.digitize` will not raise a `ValueError` when the sequence contains invalid characters for the alphabet (previously was an `UnexpectedError`).

Added
- `TraceAligner`, `Trace` and `Traces` classes to `pyhmmer.plan7` to get tracebacks after aligning several sequences against an HMM.
- `pyhmmer.hmmalign` function with the same features as the `hmmalign` binary from HMMER3.
- Support for out-of-band pickling in `easel.Vector` and `easel.Matrix`.

Changed
- Allow creating an empty `Vector` or `Matrix` by calling their constructor without arguments.

Fixed
- Potential unreported exceptions in `plan7.OptimizedProfile.write` and several `plan7.SSIWriter` methods.

Added
- `pickle` protocol for `easel.Alphabet`, `easel.Bitfield`, `easel.KeyHash`, `easel.Vector`, `easel.Matrix` and `plan7.HMM`.
- `taxonomy_id` and `residue_markups` properties to `easel.Sequence`.
- `sum_score` property to `plan7.Hit`.
- `plan7.EvalueParameters` class to expose the e-value parameters of a `plan7.HMM` or a `plan7.Profile`.
- Equality checks and slicing for `easel.Matrix` and `easel.Vector`.
- Support for creating and manipulating zero-sized `easel` matrices and vectors.
- `plan7.Cutoffs` class to expose the Pfam score cutoffs of a `plan7.HMM` or a `plan7.Profile`.
- Keyword arguments to configure E-value thresholds when creating a `plan7.Pipeline` object.
- Support for using model-specific thresholding options in `plan7.Pipeline`.

Changed
- Use the *replace* error handler when decoding error messages to skip potential decoding issues when already building an exception.
- Improve `pyhmmer.hmmer` to ensure background threads exit on a `KeyboardInterrupt`.
- `easel.VectorU8.__eq__` accepts any object implementing the buffer protocol.
- `plan7.HMM.creation_time` now takes and returns a `datetime.datetime` object, assuming the field is only ever set with `asctime`.
- Refactor `easel.Vector` and `easel.Matrix` and mark exposed memory as C-contiguous.

Fixed
- `easel.Alphabet` not reporting potential allocation errors.
- Potential buffer overflow in `easel.Matrix` and `easel.Vector` when calling `__init__` more than once.

Added
- `OptimizedProfile.convert` method to configure an optimized profile from a `Profile` without reallocating a new `P7_OPROFILE` struct.

Changed
- Rewrite the `plan7.Pipeline` search loop to avoid reacquiring the GIL between reference sequences.
- Require the reference sequences to be stored in a collection (instead of an iterable) when passing them to the `search_hmm`, `search_msa` and `search_seq` methods of `plan7.Pipeline`.
- Avoid reallocating a new `OptimizedProfile` every time a new HMM is passed to `Pipeline.search_hmm`.
- Relax the GIL while sorting and thresholding `TopHits` in `Pipeline` search methods.

Added
- `ignore_gaps` parameter to `pyhmmer.plan7.SequenceFile`, allowing to skip the gap characters when reading a sequence from an ungapped format.
- `__sizeof__` implementation for some
- Dedicated check for sequence length before running the platform-specific code in `pyhmmer.plan7.Pipeline`.

Fixed
- Score system not being set in `pyhmmer.plan7.Builder.build_msa`.
- Alphabet not being checked after the first sequence in `Pipeline` search and scan methods.