Fix provider options when several providers are passed

[fxmarty]

When several providers are passed to the InferenceSession, which is the case when TensorrtExecutionProvider is chosen, the provider_options argument needs to be of the same length than providers, otherwise raising:

EP Error using ['TensorrtExecutionProvider', 'CUDAExecutionProvider']
Falling back to ['CUDAExecutionProvider', 'CPUExecutionProvider'] and retrying.

Reference: https://onnxruntime.ai/docs/api/python/api_summary.html#inferencesession

This was untested up to now. Still need to add a test for this PR.

In a next PR: remove all the code duplication for load_model() in modeling_ort.py, modeling_decoder.py, modeling_seq2seq.py. But I won't do it in this PR.

This should fix #606

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[regisss]

LGTM!

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.