[TuplePointsToAnalysis] Be less conservative on loop fusion nodes when reusing buffer.

Previously, we say we cannot reuse operand buffer for a loop fusion
node if any of the fusion's inputs is a broadcast or reshape. That's
too conservative since in theory we can still reuse the operand's
buffer if all the users of that particular operand are elementwise.
This CL implements that. Allow sharding operand and output buffer for
partially elementwise fusions.

The same change have been recently applyed to DataFlowAnalysis as well
but we use this pass in many places as well.

PiperOrigin-RevId: 200028414