Make default codegen behavior skip Lower function (#75267)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/75267

- clean up arguments relating to ts backend generation
- make entire lowering function rather than just body be a part of
  backend-IR class

Test Plan: Imported from OSS

Reviewed By: bdhirsh

Differential Revision: D35411212

Pulled By: wconstab

fbshipit-source-id: 44419e42f706afeb967f704649c2b44e9f66d969
(cherry picked from commit 80a6fa715db97deb056db31e28689dd86a50a4bb)

tools/setup_helpers/generate_code.py

@@ -197,16 +197,17 @@ def main() -> None:
         assert os.path.isfile(ts_backend_yaml), f"Unable to access ts_backend_yaml: {ts_backend_yaml}"
         assert os.path.isfile(ts_native_functions), f"Unable to access {ts_native_functions}"
         from tools.codegen.gen_lazy_tensor import run_gen_lazy_tensor
+        from tools.codegen.dest.lazy_ir import TSLazyIR
         run_gen_lazy_tensor(aten_path=aten_path,
                             source_yaml=ts_backend_yaml,
                             backend_name="TorchScript",
                             output_dir=lazy_install_dir,
                             dry_run=False,
                             impl_path=ts_native_functions,
-                            gen_ts_lowerings=True,
                             node_base="TsNode",
                             node_base_hdr=ts_node_base,
                             build_in_tree=True,
+                            lazy_ir_cls=TSLazyIR,
                             per_operator_headers=options.per_operator_headers)