fix: resolve SQLite embedding TypeError by using sa_column=False

[octo-patch]

Fixes #382

Problem

When running the SQLite test script (tests/test_sqlite.py), table creation fails with:

raise ValueError(f"{type_} has no matching SQLAlchemy type")
ValueError: <class 'list'> has no matching SQLAlchemy type

SQLiteResourceModel, SQLiteMemoryItemModel, and SQLiteMemoryCategoryModel all inherit embedding: list[float] | None from their respective parent base classes (Resource, MemoryItem, MemoryCategory). SQLModel's metaclass traverses the full MRO to discover annotated fields and attempts to create a SQLAlchemy column for every one of them. Since SQLAlchemy has no native mapping for list[float], it raises a ValueError.

The previous @property approach tried to shadow the inherited field, but Python properties do not prevent Pydantic v2 / SQLModel's metaclass from processing the annotation found in the parent class hierarchy.

Solution

Re-declare embedding in each SQLite model class with Field(default=None, sa_column=False). This tells SQLModel the field exists for Pydantic validation purposes only — no SQL column should be created for it.

The actual persistence of embeddings is unchanged: embedding_json (a Text column) stores the vector serialized as a JSON string, and the repository layer's _normalize_embedding / _prepare_embedding helpers handle conversion between list[float] and JSON on every read/write.

The now-unreachable @property / @embedding.setter bodies and their associated import json / import logging are also removed.

Testing

• SQLite table creation no longer raises ValueError: <class 'list'> has no matching SQLAlchemy type
• Read/write round-trips for embeddings are unaffected because all repository code already accessed embedding_json directly, not the property