🧞 Fix genai exporter arguments bug and add Genai phi2 example

[trajepl]

Describe your changes

1. With the latest onnxruntime-genai, there are a few bugs:
   a. precision can only accept string value(int4, fp16, fp32) but not the
   
   b. execution_provider accept ep value but not device value
   c. the pass of GenAIModelExporter is not EP agnostic. For cuda int4, the input shape is changed to fp16, but for cpu int4, it is still fp32.

2. Add phi2 example.

• CUDA: v100
  
• CPU: Intel(R) Xeon(R) CPU E5-2698
  

4. Update genai generation example in README of llama2/phi2.

Checklist before requesting a review

• Add unit tests for this change.
• Make sure all tests can pass.
• Update documents if necessary.
• Lint and apply fixes to your code by running lintrunner -a
• Is this a user-facing change? If yes, give a description of this change to be included in the release notes.
• Is this PR including examples changes? If yes, please remember to update example documentation in a follow-up PR.

(Optional) Issue link

[guotuofeng]

    INT4 = "int4"

I think add __str__ could fix the str(Precision.FP32) issue.

---

Refers to: olive/passes/onnx/genai_model_exporter.py:31 in b5f1e71. [](commit_id = b5f1e71, deletion_comment = True)

[guotuofeng]

Basically, hardware dependent means need create InferenceSession and need run on target.

[jambayk]

Yes, like Mike says, a search point can be invalid for a particular ep while the pass is accelerator agnostic.

Pass is only non-agnostic if the output of the pass changes based on the accelerator.

[jambayk]

Actually, line 81 depends on the accelerator_spec so it satisfies the requirement.

[trajepl]

Seems, we need an extra pass config like target_ep. Then the is_accelerator_agnostic can be True.
Synced with Mike, the accelerator_spec here is target device so current change is fine.

For genai exporter, the output will be different for different ep.
For cuda int4, the input shape is changed to fp16, but for cpu int4, it is still fp32.

[guotuofeng]

    INT4 = "int4"

I think add __str__ could fix the str(Precision.FP32) issue.

Refers to: olive/passes/onnx/genai_model_exporter.py:31 in b5f1e71. [](commit_id = b5f1e71, deletion_comment = True)

Please make sure the search point is always Enum. If there are any possibility to use str for precision, we'd better add __str__