[tritonbench] A better way to encode input shapes with parameters

Right now developers need to manually specify the input shape metadata to generate the input tensors.
I am thinking if we could use decorators to make this process easier and more user-friendly.

Old style:

def get_input_iter(self) -> Generator:
    shape_a = range(x, y)
    shape_b = range(m, n)
    for a in shape_a:
        for b in shape_b:
            yield generate_input(a, b)

New style:

@given(
        a=st.integers(min_value=x, max_value=y),
        b=st.integers(min_value=m, max_value=n),
)
def get_input(self, a, b):
      return generate_input(a, b)

The issue does not name files, tests, or an entry point. Start by locating tritonbench's input-shape metadata and tensor-generation code, then compare the current manual iterator flow with the proposed decorator-based interface. Done means developers can express parameterized input shapes without manually writing the nested iteration.