sample() method should accept very large num_reads value (not limited by underlying HW solver)

**Current Problem**
Related to https://github.com/dwavesystems/dwavebinarycsp/issues/57, the need to obtain an inordinately large number of samples meant that the straightforward problem statement, to request a large number of samples, had to be turned into a while() loop (lines 46-62 in attached code)
[maxCut_csp.py.gz](https://github.com/dwavesystems/dwave-system/files/2402961/maxCut_csp.py.gz)
because the EmbeddingComposite.sample() method is limited to 10,000 samples, which was not enough to find a valid sample. The higher-level sample() method should accept a large value (at least millions) and, if necessary, strip-mine that number to comply with lower-level sampler constraints.