Dwave-hybrid

New Features
---

- Unified SDK docs

Fixes
---

- Python 3.8 support
- Loosened NumPy version requirements

New Features
---

- `ICM`/`IsoenergeticClusterMove` samples processor block
- `ExplodeSamples` block that splits sampleset samples into states
- `HybridSampler` can now return the final state produced by the
workflow, and it can propagate sampling parameters as workflow
run-time parameters

Fixes
---

- Properly handle list samples in `utils.flip_energy_gains`

New Features
---

- `EnergyWeightedResampler` for input set re-sampling at a given
temperature
- `AggregatedSamples` block that aggregates or spreads input samples
- `Const` block that sets output state variables to constant values
- `Wait` block for synchronization of racing components
- Population annealing workflow examples and importable workflows
(`PopulationAnnealing`/`HybridizedPopulationAnnealing`)
- Input extrapolation and output aggregation options in
`FixedTemperatureSampler`
- `auto_embedding_params` argument in QPU auto-embedding samplers
- `Loop` now terminates on `EndOfStream` (alternative to `Unwind`)

Fixes
---

- All QPU samplers now explicitly ask for a QPU solver
- Faster and more general `SplatComposer`/`hstack_samplesets` (act on
all samples)
- Support for unstructured QPU samplers in QPU auto-embedding samplers
- Isolated RNG in PT/PA components

Changes
---

- Renamed `qpu_params` argument to `sampling_params` in all QPU samplers
- New default for `rolling_history` in `EnergyImpactDecomposer`: now
decomposing the complete problem
- Removed default variables (`samples` and `problem`) from `State`
- `Identity` block is now non-blocking. In racing contexts you should
now use `BlockingIdentity` (or `Wait`) instead.

New Features
---

- Added `Branches` and `Dup` components that generalize `Parallel`
- Added predicate-based termination criterion in `Loop`, enabling run-to-energy mode
- Even lazier auto-embedding QPU sampler
- Added parallel tempering workflow examples
- Importable workflows for `Kerberos`, `ParallelTempering`/`HybridizedParallelTempering`
and `SimplifiedQbsolv`

Fixes
---

- Looser traits enforcing (run-time only, skip construction-time checks)
- `RandomConstraintDecomposer` edge cases
- `RoofDualityComposite` energy calculation

New Features
---

- Tabu/SimulatedAnnealing samplers now use input state samples to initialize
- Racing context propagated to children via runopts
- Added `stoppable` class decorator to simplify implementing thread-safe stop behaviour
- Added `RoofDualifyDecomposer`
- Exposed all SA parameters in SimulatedAnnealing sampler
- Exposed non-interruptible Tabu timeout in Kerberos

Fixes
---

- Longer default runtime in Tabu
- Better handling of larger problems by Kerberos

Changes
---

- `endomorphic` option removed from `RacingBranches` and `ParallelBranches` in favor of explicit `Identity` block
- `sweeps` renamed to `num_sweeps` in SA samplers