This requires using dtype strategies from hh instead of xps.
This also fixes some functions in linalg and fft that were incorrectly only tested against real floating-point dtypes, due to Hypothesis's confusing nomenclature of "floating_dtypes" being only real floating-point dtypes (and this also eliminates the use of Hypothesis's confusing "scalar_dtypes").
This requires using dtype strategies from hh instead of xps.
This also fixes some functions in linalg and fft that were incorrectly only tested against real floating-point dtypes, due to Hypothesis's confusing nomenclature of "floating_dtypes" being only real floating-point dtypes (and this also eliminates the use of Hypothesis's confusing "scalar_dtypes").
Fixes #265