Remove TENSOR_BOOL8 support from comparison ops

Ops changed: LESS, LESS_EQUAL, GREATER, GREATER_EQUAL.

Support for this was added in Q but there are no actual use cases for
it. There were also no tests covering this path so we decided to remove
it.

Higher level frameworks such as TF Lite and ONNX do not support
comparisons of booleans either. Hardware manufacturers will have less
things to implement after this change.

Test: NeuralNetworksTest_static
Bug: 129570421
Change-Id: Ic744db00d5401633853e474e6b80bb735d1d3f8e