Fix crashes in multi-column detection on position-less shapes

[eeshsaxena]

With --try-multi-column, converting a deck that contains a shape without an explicit position crashes.

python-pptx returns None for shape.top/left/width when the shape's XML has no offset (real .pptx files do this). multi_column.py then hits two crashes:

# is_two_column_text / assign_shapes:
sorted(ungroup_shapes(slide.shapes), key=attrgetter('top', 'left'))
# TypeError: '<' not supported between instances of 'NoneType' and 'Emu'

centroid_x = shape.left + shape.width / 2
# TypeError: unsupported operand type(s) for +: 'NoneType' and 'float'

Fixes:

• Sort with a None-safe key (lambda s: (s.top or 0, s.left or 0)).
• Skip shapes without a position/size when computing centroids (they have no horizontal centroid to contribute).
• raise (ValueError, "...") (3 places) raises a tuple, which Python 3 rejects with TypeError: exceptions must derive from BaseException; changed to raise ValueError("...").

Added a regression test that builds a slide mixing positioned shapes with a position-less one; it crashes on master and passes with this change.

[eeshsaxena]

Hi! Gentle nudge on this one whenever you have some bandwidth. It's a small, self-contained fix (Fix crashes in multi-column detection on position-less shapes), and it's currently mergeable with no conflicts. No urgency at all, and I'm happy to make any changes you'd like. Thanks for maintaining pptx2md!