[dynamo] Add "Torch-Compiled Region" profiler event (#108462)
**Motivation**: We already have a `CompiledFunction` event that comes from the autograd.Function added by aot_autograd. However, this doesn't appear during inference, or if none of the inputs to a graph require grad. It also doesn't appear if your backend doesn't use aot_autograd.
This adds a profiler event that will always appear.
<img width="615" alt="Screenshot 2023-09-01 at 4 46 28 PM" src="https://github.com/pytorch/pytorch/assets/5067123/fed90ca9-a8e7-458c-80eb-b4160de55218">
Perf - increase in latency (with profiler turned off) was within noise when I measured a simple cpu-only torch-compiled function that returned `x.view_as(x)`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/108462
Approved by: https://github.com/anijain2305