Dwave-tabu

Fixes
---

- Modernized setup and build pipeline (85)

- Package metadata moved to setup.cfg
- Use PEP-517/PEP-518 build system via pyproject.toml
- Dropped 32-bit/i686 manylinux wheels
- Added manylinux2014 wheels
- Dropped AppVeyor in favor of CircleCI
- Synchronized deploys
- New CI workflow (build dist -> install from dist -> test)
- Simplified and standardized CI by using [`dwave/ocean` orb](https://circleci.com/developer/orbs/orb/dwave/ocean)

New Features
---

- Biases now stored as double floats, instead of as ints (29)

- Added C++ code tests

- Added `seed` parameter to `TabuSampler.sample()` for reproducible
random initial states (30)

- Added `energy_threshold` to `TabuSampler.sample()` to enable target energy
based termination (48)

Fixes
---

- C++ code refactored and modernized

- `TabuSampler` now implements the full `dimod.Initialized` interface

- RNG isolated per instance and thread (71)

Changes
---

- Removed bias scaling (`scale_factor` from `TabuSampler.sample()`
args), since biases are now internally represented as
double-precision floats (29)

- Switched to Cython from SWIG for C++/Python interface

- Dropped Python 3.5 support

- Removed `init_solutions` argument of `.sample()` (deprecated, since v0.2.0,
in favor of `initial_states`).

New Features
---

- Biases stored as double floats, instead of as ints (29)

- Added C++ code tests

- Added `seed` parameter to `TabuSampler.sample()` for reproducible
random initial states (30)

- Added `energy_threshold` to `TabuSampler.sample()` to enable energy
target based termination (48)

Fixes
---

- C++ code refactored and modernized

- `TabuSampler` now implements the full `dimod.Initialized` interface

- RNG isolated per instance and thread (71)

Changes
---

- Removed bias scaling (`scale_factor` from `TabuSampler.sample()`
args), since biases are now internally represented as
double-precision floats (29)

- Switched to Cython from SWIG for C++/Python interface

- Dropped Python 3.5 support

Fixes
---

- Publish Python 3.9 wheels for Linux/macOS/Windows

New Features
---

- dimod 0.9.x support (all BQM types)

Changes
---

- Dropped Python 2 and 3.4 support