Eager mode for onnxscript

[justinchuby]

1. Create an eager graph building mode
2. The mode needs to be eager, aka the graph is built on the fly, instead of built lazily (?)
3. Types for ONNX Script functions should pass mypy checks. This is important for adoption in other code bases because we don't want our users to disable their checks for onnx script
4. We can create overloaded methods to handle numpy (concrete value) inputs. The inputs can also be ir.Tensor. Users should annotate the functions to take ir.Value or a derivative of this class (aka. onnxscript Tensor)
5. Typing annotations should be simplified because parsing annotations and keeping up with different python versions is a pain. Instead, decorate the functions with onnxscript decorators and declare type information there, if needed.

[justinchuby]

Done