feat: add semantic embedding support to MemoryRecallPipeline

[BiswasNehaa]

Closes #7

What

• Implements _semantic_score() using sentence-transformers (optional dependency)
• Stores embeddings in Memory.embedding on creation via MemoryWriter.create_candidate
• Falls back gracefully to None when sentence-transformers isn't installed or no embedding exists
• Adds sentence-transformers as an optional embeddings extra in pyproject.toml
• Adds a regression test verifying graceful fallback behavior

Testing

Ran python -m pytest tests/test_memory_lifecycle.py -v — both tests pass.