[Documentation] Optimizations and dependence on hardware

### Describe the documentation issue

Am I correct in assuming that since extended optimizations are applied to nodes with a certain EP, then they are dependent on hardware?

So, if I'm understanding correctly, does that mean extended optimizations are best applied at runtime, i.e. in this case only in online mode, and offline mode is not available, because there may be performance problems?

Thanks in advance.

### Page / URL

https://onnxruntime.ai/docs/performance/model-optimizations/graph-optimizations.html#extended-graph-optimizations

Start with the extended graph optimizations section at the linked graph-optimizations.html page and review how execution providers and online/offline optimization are described there. Clarify whether the optimizations depend on hardware or the selected provider, and document the correct runtime guidance so the page answers both questions.

Written by the indexing model from the issue text.