[menny]

Summary

Implements Pipeline 1 (Build-Time seed.db Generation) as specified in docs/INGESTION.md:

• Step 1: Kaikki Dictionary Streaming: Streams Kaikki Spanish JSONL line-by-line, resolves SUBTLEX-ESP frequency rankings, detects cognates using CognateDetector, and expands base and inflected forms into dictionary.
• Step 3: False Cognates Expansion: Expands trap seeds across all conjugated and plural forms in the dictionary into false_cognates.
• Step 4: Tatoeba Processing: Reads bilingual sentence pairs, scores them with SentenceScorer, and packages valid sentences into curated_lessons.
• Cognate Rules Seeding: Populates cognate_rules from JSON definitions.
• SQLite Database Management: Transaction-safe batched persistence with schema initialization and indexes.
• Preprocessor CLI & Tests: //tools/preprocessor binary entry point and end-to-end integration test suite.

Stacked on top of #6 (sentence-scorer).