[dynamo, nested graph breaks] fix nested step graph break related issues (#162737)

Turns out codegen'ing a nested step graph break is significantly more complicated than first thought. The optimized function should actually do:
- call graph/load values/do side effects etc.
- call into the leaf's resume function, but skipped (this essentially step graph break function for just the leaf function)
- call into all the other resume functions, traced.

This PR also adds `torch._dynamo.step_unsupported()`, which can be used for internal testing purposes to better test step graph break handling.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/162737
Approved by: https://github.com/Lucaskabela
ghstack dependencies: #160601

torch/_dynamo/exc.py

@@ -263,6 +263,11 @@ class RecompileLimitExceeded(Unsupported):
     pass
 
 
+# debug exception thrown when tracing torch._dynamo.step_unsupported()
+class StepUnsupported(TorchDynamoException):
+    pass
+
+
 class UnsafeScriptObjectError(TorchDynamoException):
     pass