Add neuronx cache registry

[dacorvo]

This extends the neuronx caching for decoder models to store the cached neuron configurations under a specific registry folder for each neuronxcc compiler version.

aws-neuron/optimum-neuron-cache/                                                
└── neuronxcc-2.12.54.0+f631c2365
    └── registry
       ├── hf-internal-testing
       │   └── tiny-random-gpt2
       │       └── 8019d93dd8eda6d8da82.json
       └── mistralai
            └── Mistral-7B-Instruct-v0.1
                └── b12f53f65aebf02fdfa9.json

This also adds a get_hub_cached_entries helper and CLI command to lookup cached configurations for a specific model_id.

The output looks like this:

$ optimum-cli neuron cache lookup mistralai/Mistral-7B-Instruct-v0.1
*** 1 entrie(s) found in cache for mistralai/Mistral-7B-Instruct-v0.1 ***


task: text-generation
batch_size: 1
num_cores: 2
auto_cast_type: bf16
sequence_length: 2048
compiler_type: neuronx-cc
compiler_version: 2.12.54.0+f631c2365
checkpoint_id: mistralai/Mistral-7B-Instruct-v0.1
checkpoint_revision: 9ab9e76e2b09f9f29ea2d56aa5bd139e4445c59e

[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.

[michaelbenayoun]

Is it just formatting?
Anyways those example are automatically generated from the Transformers library so this change will be overriden when we synchronize with the next release.

[dacorvo]

If I don't apply them, then the check-code-quality fails. I could exclude the examples from it though ...

[michaelbenayoun]

No worries, I just wanted to check if the change was only about styling. After reading the rest of the PR I understood it was automatic reformatting.

[michaelbenayoun]

LGTM thanks!