Rename a feature inside nested MIL blocks

[LeSingh1]

rename_feature's ML Program branch walks each function's block_specializations but not the nested blocks that control flow operations hold. cond and while_loop blocks reference outer vars by name, so renaming an input or output that they use leaves those references pointing at the old name. The renamed model then fails to load:

Unable to parse ML Program: at unknown location:
Input 'b' for parameter 'x' does not resolve.

Repro on main:

@mb.program(input_specs=[mb.TensorSpec(shape=(1,), dtype=types.bool),
                         mb.TensorSpec(shape=(1,))])
def prog(a, b):
    def true_fn():
        return mb.add(x=b, y=1.0)
    def false_fn():
        return mb.mul(x=b, y=2.0)
    return mb.cond(pred=mb.squeeze(x=a), _true_fn=true_fn, _false_fn=false_fn)

model = ct.convert(prog, convert_to="mlprogram",
                   minimum_deployment_target=ct.target.iOS16)
spec = model.get_spec()
ct.utils.rename_feature(spec, "b", "renamed_b")
ct.models.MLModel(spec, weights_dir=model.weights_dir).predict(...)  # raises

The block walk is now a recursive helper so nested blocks get the same treatment. Added a test that renames an input used by both cond branches and checks the predictions on each branch; it fails on main with the error above.