Remove onnx dependency

[echarlaix]

• add base configs classes like OpenVINOConfig and TextEncoderOpenVINOConfig that were originally defined in https://github.com/huggingface/optimum-onnx/blob/main/optimum/exporters/onnx/config.py
• move all XxxInputGenerator classes in input_generator.py
• remove deprecated export_pytorch_via_onnx (previously used when we were converting the model to ONNX first and then converting it to OV IR)
• remove opset argument (previously needed for ONNX export)

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[IlyasMoutawwakil]

Tremendous work 🙏 LGTM !

[rkazants]

@echarlaix, can you please double check these failures https://github.com/huggingface/optimum-intel/actions/runs/26757786426/job/78913206681?pr=1753:

FAILED tests/openvino/test_decoder.py::OVModelForCausalLMIntegrationTest::test_pipeline_44_phimoe - AttributeError: 'PreTrainedConfig' object has no attribute 'max_position_embeddings'
FAILED tests/openvino/test_decoder.py::OVModelForCausalLMIntegrationTest::test_pipeline_51_gpt_oss - AttributeError: 'PreTrainedConfig' object has no attribute 'max_position_embeddings'
FAILED tests/openvino/test_decoder.py::OVModelForCausalLMIntegrationTest::test_pipeline_52_gpt_oss_mxfp4 - AttributeError: 'PreTrainedConfig' object has no attribute 'max_position_embeddings'
FAILED tests/openvino/test_decoder.py::OVModelForCausalLMIntegrationTest::test_pipeline_53_afmoe - TypeError: 'AfmoeExperts' object is not subscriptable

[echarlaix]

@echarlaix, can you please double check these failures https://github.com/huggingface/optimum-intel/actions/runs/26757786426/job/78913206681?pr=1753:

FAILED tests/openvino/test_decoder.py::OVModelForCausalLMIntegrationTest::test_pipeline_44_phimoe - AttributeError: 'PreTrainedConfig' object has no attribute 'max_position_embeddings' FAILED tests/openvino/test_decoder.py::OVModelForCausalLMIntegrationTest::test_pipeline_51_gpt_oss - AttributeError: 'PreTrainedConfig' object has no attribute 'max_position_embeddings' FAILED tests/openvino/test_decoder.py::OVModelForCausalLMIntegrationTest::test_pipeline_52_gpt_oss_mxfp4 - AttributeError: 'PreTrainedConfig' object has no attribute 'max_position_embeddings' FAILED tests/openvino/test_decoder.py::OVModelForCausalLMIntegrationTest::test_pipeline_53_afmoe - TypeError: 'AfmoeExperts' object is not subscriptable

@rkazants the failing tests you shared are related to transformers main (not related to this PR but from incompatibility between optimum-intel and transformers main), let me know if you're referring to something else