Reintroduce Face Classification

[derneuere]

This pull request reintroduces classification alongside the existing clustering functionality, ensuring both methods work together seamlessly. It also implements a soft delete feature, improving data handling and user control. Additionally, the refactor enhances clarity in differentiating user input from algorithmic input, addressing ambiguities in data origins.

Key Changes

Classification Reintroduction:
Adds classification without disrupting the current clustering setup.
Classification improves accuracy for larger datasets where most persons are already known.

Soft Delete Feature:
Allows entities to be "soft deleted," enabling recovery and improving data integrity without permanent deletion.

Input Differentiation:
Replaces the binary inferred field with three distinct relationships to persons. These relationships clarify whether the input was matched by a user or one of two algorithms.
Enables tracking of which algorithm matched an entity, providing transparency and control over the results.

Addressed Issues
This PR resolves multiple issues related to input ambiguity and clustering limitations, including: #978 #848 #1409 #879
#600 #598 #501

Rationale Behind Clustering and Classification

Clustering was retained due to its effectiveness in cold start scenarios where no user data exists. It helps match faces even when no persons have been added to the system.

However, common clustering problems include:

• Inability to restrict face matching to known persons.
• Lack of improvement from prior user input.

Classification addresses these limitations by improving accuracy in cases where all persons are already known, ensuring better results for large datasets.

Additional Improvements

• Ability to identify which algorithm found a match and which did not.
• Users can now switch between algorithmic results, giving more control over the matching process.

[derneuere]

#1458

[derneuere]

#458