[Parsiffall1]

Fixes #10358.

Changes proposed in this pull request:

• PyPath_Flatten() uses the length from PyObject_Length(), which calls a subclass's __len__. For tuples it then read items with the unchecked PyTuple_GET_ITEM(), so a tuple subclass that reported a larger length read past the end of the tuple and crashed. This uses PyTuple_GetItem() instead, which raises IndexError, matching what already happens for list subclasses.
• Adds a test with a tuple subclass whose __len__ is larger than its contents.

[radarhere]

Note that I'm not sure if the issue needs to be fixed.

[akx]

Same class of issue as #10198, #10055. PyPath_Flatten is also presumably then also prone to crash if a member of the sequence mutates the sequence (since assign_item_to_array can end up calling back to user coercion code) (which is the thing #10055 tries to fix).