BUG: raise clear ValueError for non-Polygon geometries in MultiPolygon

What

MultiPolygon.__new__ only type-checked the first element of its input
list. A non-Polygon BaseGeometry (for example a GeometryCollection or
Point) appearing after the first element fell through to the
(shell, holes) unpacking path and hit shell = ob[0], raising a
cryptic error:

import shapely
poly = shapely.box(0, 0, 1, 1)
shapely.MultiPolygon([poly, shapely.GeometryCollection([poly])])
# TypeError: 'GeometryCollection' object is not subscriptable

This change detects BaseGeometry instances that are not Polygons in
the construction loop and raises a ValueError naming the offending
geometry type instead:

ValueError: Input must be valid Polygon objects or sequences of
(shell, holes) tuples, got a GeometryCollection

Why

The previous behaviour surfaced an implementation detail (__getitem__)
rather than the actual problem (an invalid member type). The new check
mirrors the existing member-type guard that already rejects
MultiPolygon inputs, keeping the constructor's error reporting
consistent.

Valid inputs are unaffected: plain Polygon objects, (shell, holes)
sequence/tuple inputs, and construction from an existing MultiPolygon
all continue to work. MultiPolygon members are still caught earlier by
the existing dedicated check.

Tests

Added test_fail_list_with_non_polygon_geometry (mirroring
test_fail_list_of_multipolygons), parametrized over GeometryCollection,
Point, LineString, and LinearRing. Each type is checked in three
positions: as the only element, before a valid Polygon, and after a valid
Polygon. Every case asserts a ValueError naming the offending type.
Verified the new cases fail on the unpatched source (the original
TypeError) and pass with the fix. The full shapely/tests/geometry suite
passes.

Closes #2178