Add complex number support to linspace

[kgryte]

This PR

• adds complex number support to linspace.
• allows start and stop to be complex numbers.
• allows dtype to a complex floating-point data type, not just a real-valued floating-point data type.
• specifies that spacing between complex numbers be computed by independently computing the spacing for real and imaginary components, respectively. This follows existing practice in NumPy, PyTorch, Julia, MATLAB, and elsewhere.
• specifies that if dtype is None, if either start or stop is a complex number, then the output dtype must be the default complex floating-point dtype; otherwise, the output dtype must be the default real-valued floating-point dtype.
• adds a warning that conversions of start and stop should follow type promotion rules; conversions not following type promotion rules are implementation-dependent (e.g., NumPy allows specifying a real-valued dtype when start and/or stop is complex by discarding the imaginary component; as such a conversion is not specified in this standard, such behavior is implementation-defined).
• removes a requirement that linspace raise an exception if num is a not a nonnegative integer, as the standard typically avoids specifying validation behavior, leaving it up to conforming libraries to determine how to best handle invalid inputs (e.g., raise an exception, clip inputs, etc).

[rgommers]

This all looks correct, in it goes. Thanks @kgryte