[PT2] Provide public API to get compile options/configs

[yanboliang]

[From Meta internal use case]
For inductor, we have two ways to set up config:

• Set up via torch._inductor.config.triton.cudagraphs = True.
• Passing in options when calling torch.compile(model, options={"triton.cudagraphs": True}).

There is requirement that we would like to get torch.compile config for further usage. E.g, the training platform wants to get if model is compiled with cudagraph enabled and trigger further action. But currently the configs from the above two sources are merged internally, there is no public API to access this.

cc @ezyang @msaroufim @wconstab @ngimel @bdhirsh @anijain2305

[williamwen42]

@qihqi's dataclass config PR may impact this: #103041.