feat(queries): add classification and metadata to saved queries

[shefeek-jinnah]

Summary

• Add automatic SQL query classification that analyzes DataFusion logical plans and categorizes saved queries into: full_scan, projection, filtered_scan, point_lookup, aggregation, join
• Store rich classification metadata per version: category, num_tables, has_predicate, has_join, has_aggregation, has_group_by, has_order_by, has_limit
• Add user-provided tags (array) and description (text) fields to saved queries
• Add category_override and table_size_override columns for future manual overrides (read APIs use COALESCE(category_override, category))
• Add v7 migrations for both SQLite and Postgres

API Changes

• POST /v1/queries — accepts optional tags and description; response includes classification fields
• PUT /v1/queries/{id} — accepts optional tags and description; updates without creating new version when only metadata changes
• GET /v1/queries/{id} and GET /v1/queries — responses include tags, description, and classification fields
• GET /v1/queries/{id}/versions — version entries include classification fields
• Classification fields are omitted from JSON when null (e.g., if plan analysis fails)
• Tags are trimmed and validated (max 50 tags, max 100 chars each; whitespace-only rejected)
• Description validated (max 10,000 chars)

Design Decisions

• Classification is best-effort: if logical plan building fails (e.g., missing tables), fields are stored as NULL and a warning is logged
• Classification uses a fresh SessionContext per call to avoid racing with concurrent queries
• Tags stored as JSON text in a single column for simplicity
• Tags JSON deserialization failures log a warning and fall back to empty (consistent with timestamp parsing pattern)