Add Feature Extraction Support for API Classifiers

[mohamedelabbas1996]

Description

This PR adds support for returning model feature vectors (embeddings) alongside classification results in the Data Companion API.
The classification pipeline now supports returning a vector embedding per classification, derived from the classification model backbone.

The changes are fully backward-compatible for models that do not implement custom get_features(), as they will fallback to returningNone from the base class.

Related Issues

#752

Screenshots

Detection features clustering visualization using K-means + PCA

[sentry]

🔍 Existing Issues For Review

Your pull request is modifying functions with the following pre-existing issues:

📄 File: trapdata/api/models/classification.py

Function	Unhandled Issue
save_results	ValidationError: 15 validation errors for ClassificationResponse ... Event Count: 2
save_results	ValidationError: 10 validation errors for ClassificationResponse ... Event Count: 2
save_results	AttributeError: 'NoneType' object has no attribute 'tolist' ... Event Count: 1
save_results	ValueError: not enough values to unpack (expected 3, got 2) ... Event Count: 1

---

_{Did you find this useful? React with a 👍 or 👎}

[mihow]

I think we should make these optional dependencies and just use numpy in the tests. unless we need to use them in the core app.

[tool.poetry.extras]
dev = ["plotly", "scikit-learn"]

[mihow]

Nice work on this method of extracting features! It seems more flexible than our current feature extractor. Perhaps we should add a comment in both feature extractors that the other one exists. And eventually update the old one to use this code.