fix: accept zero WordEmbeddingDistance thresholds

[emecii]

What does this PR do?

Summary

This fixes WordEmbeddingDistance so that 0.0 is treated as an explicitly selected threshold rather than as an omitted value.

Additions

• Regression coverage using a small in-memory embedding.

Changes

• Use explicit None checks when validating the mutually exclusive min_cos_sim and max_mse_dist options.
• Apply both selected metrics when their threshold is 0.0.

0.0 is a meaningful boundary: cosine similarity must be non-negative, and MSE distance must be zero. The previous truthiness checks rejected those constructor calls and would have skipped enforcement if accepted.

Deletions

• None.

Validation

• python -m pytest tests/test_word_embedding.py -k word_embedding_distance -q — 5 passed
• python -m pytest tests/test_constraints -q — 8 passed
• make lint — passed

The full tests/test_word_embedding.py run was not used as the acceptance check because an existing network-backed embedding test stalled while downloading its artifact; the deterministic selected tests completed normally.

Checklist

• The title summarizes the contribution.
• This is ready for review, not a draft.
• Relevant tests were added and passed.
• Existing relevant constraint tests passed.
• No new public methods or modules were added.