xref https://github.com/Toblerity/Shapely/pull/1180/files#r693378882

PyGEOS geometries are "immutable", not only regarding the coordinates of the geometry, but also the python object: you can't assign random attributes as you can do with a normal class defined in python (this is because they are C extension types):

>>> import pygeos
>>> p = pygeos.Geometry("POINT(1 1)")
>>> p
<pygeos.Geometry POINT (1 1)>
>>> p.name = "test"
...
AttributeError: 'pygeos.lib.Geometry' object has no attribute 'name'

However, because in Shapely we still create a Python-defined subclass for each of the geometry types (and register this, so the C code will actually instantiate those subclasses), we get the typical behaviour of standard python classes still being "mutable" objects:

# using shapely-2.0 branch
>>> p = pygeos.Geometry("POINT(1 1)")
>>> p
<shapely.geometry.Point POINT (1 1)>
>>> p.name = "test"
>>> p.name
'test'

It's however relatively easy to keep the "can't set custom attribute" behaviour of the base C class (if we want this), by adding an empty __slots__ to each of the python subclasses. (which is what this PR is doing)

Personally, I liked the fact that pygeos Geometries were completely immutable, so I would be in favor of preserving this characteristic in Shapely as well.
(for example, we will generally re-use objects (eg when broadcasting) because they are immutable, and for such cases, setting attributes might give surprising behaviour)

If we would want to disallow setting attributes in Shapely 2.0, following PyGEOS behaviour, that's of course a breaking change compared to Shapely 1.x, but I think it should be relatively easy to add a deprecation warning for this change as well (adding a warning in __setattr__).

cc @sgillies @caspervdw