BUG: raise a clear error when cwt() is given a discrete wavelet (gh-776)

[gaoflow]

Fixes #776.

cwt(data, scales, 'coif1') (or any other discrete wavelet) crashed with:

AttributeError: 'pywt._extensions._pywt.Wavelet' object has no attribute 'complex_cwt'

The name is resolved with DiscreteContinuousWavelet, which happily returns
a discrete Wavelet for a name like 'coif1'. cwt then unconditionally
reads wavelet.complex_cwt, an attribute that only exists on
ContinuousWavelet, so the call dies with the confusing error above instead
of telling the user they passed the wrong kind of wavelet.

As @rgommers noted on the issue, passing the wrong wavelet name should give a
clear message rather than a confusing one. This validates that the resolved
wavelet is continuous and otherwise raises a ValueError that names the
offending wavelet and points to pywt.wavelist(kind='continuous'):

ValueError: cwt() requires a continuous wavelet, but 'coif1' is a discrete
wavelet. Use a continuous wavelet such as those returned by
pywt.wavelist(kind='continuous') (e.g. 'morl', 'mexh', 'cmor').

A regression test covering a name, a discrete Wavelet instance, and a
still-working continuous wavelet is included.

[rgommers]

LGTM, thanks @gaoflow.