tocomplex and c versions of PDL::Ops that have complex versions

[mohawk2]

tocomplex (no _ to match topdl) is a need that should be fulfilled. That would then allow r2C to switch to giving an error for complex inputs.

Similarly, I think that some functions should have c versions that have complex outputs. csqrt is a really good one that meets a need, separate from what I fully agree was a real need for csqrt_up. log is another. Are we basically talking about a large chunk of what PDL::Ops provides that should have "treat real as complex" versions, starting with c to distinguish them? Or just trigonometric functions?

Originally posted by @mohawk2 in #524 (comment)

[mohawk2]

@wlmb replied:

Some real functions have a bounded domain, such as log, which is
meaningless as a real function for a negative argument, since the
exponential of any real number is positive, or acos, meaningless for
arguments larger than 1 or smaller than -1, as cos of any real number
is bounded between -1 and 1. However, these functions do have a
meaning when they are interpreted as complex functions of a complex
variable where the complex variable turned out to lie on the real
line. It is for this class of functions, for which I guess a version
with starting 'c' would be useful. I can think of csqrt, clog, cacos and
casin. For example, we wouldn't need a catan, as atan is real for any
real input. We would need catanh as atanh only works in the range -1,1
and we would need cacosh as acosh only works in the range 1,Inf. I'm
not sure about other functions.