New Features
------------
* `SampleSet` accepted for `initial_states` value
* `initial_states` are truncated or expanded to match the
number of reads, and the expansion algorithm is controlled
with a new parameter, `initial_states_generator`
* Undefined `num_reads` inferred from `initial_states`
Fixes
-----
* For `initial_states` given in the 2-tuple format, infer the
vartype from the BQM
* Avoid GIL locks when interrupt function not defined
Deprecations
------------
* The 2-tuple format (NumPy array + variable labels) for
`initial_states` in favor of `SampleSet`
* `sweeps` parameter in favor of `num_sweeps`
Breaking changes
----------------
* New default value for `num_reads`: now matches `initial_states`
length, or defaults to 1