feat: add support to compute sensitivities

[JPXKQX]

Description

This PR introduces functionality for computing sensitivities within anemoi-inference. Sensitivities help quantify how model inputs influence a perturbed output, enabling better interpretability and diagnostic analysis during different forecast events.
Workflow

The sensitivity computation workflow follows a similar pattern to running a forecast:

• Select a checkpoint, an initial time, and the forecast length.
• Define a perturbation applied to the model output.
• Compute sensitivities, returned as a tensor with the same shape as the input_state.

Currently Supported

• Perturbations defined as a % of the predicted field for a single variable.
• Forecast length limited to 1.

Planned Extensions

• Additional perturbation types (e.g., random noise, location-specific, etc.).
• Support for forecast lengths greater than 1.

As a contributor to the Anemoi framework, please ensure that your changes include unit tests, updates to any affected dependencies and documentation, and have been tested in a parallel setting (i.e., with multiple GPUs). As a reviewer, you are also responsible for verifying these aspects and requesting changes if they are not adequately addressed. For guidelines about those please refer to https://anemoi.readthedocs.io/en/latest/

By opening this pull request, I affirm that all authors agree to the Contributor License Agreement.