fix: TStreamerLoop code generation should make an array

[jpivarski]

In #1275, @yixianz found that reading data with a TStreamerLoop did not raise an error (including the number of bytes check at the end of the object), but the data was a single scalar, rather than an array.

We've never seen an instance of TStreamerLoop before:

uproot5/src/uproot/streamers.py

Line 1200 in aa8b94f

# untested as of PR #629

and (my guess for) the generated code was only a little wrong:

for tmp in range(self.member('fNp')):
    self._members['fPoly'] = c('TSplinePoly3').read(chunk, cursor, af.add_to_path(forth_obj, context, 'fPoly'), file, self._file, self.concrete)

instead of

self._members['fPoly'] = numpy.empty(self.member('fNp'), dtype='O')
for tmp in range(self.member('fNp')):
    self._members['fPoly'][tmp] = c('TSplinePoly3').read(chunk, cursor, af.add_to_path(forth_obj, context, 'fPoly'), file, self._file, self.concrete)

That is, I knew that it should read as many objects as in the referred-to member (fNp), but for some reason overwrote a single object instead of making an array of objects.

This PR fixes that. @yixianz's example file is now a part of the regular test suite.