Add training workflow, datasets, and runbook

training_data/curated/dataset.stats.json

@@ -0,0 +1,9 @@
+{
+  "docs_used": 1618,
+  "chunks_written": 1778,
+  "chunk_chars": 5000,
+  "overlap_chars": 300,
+  "min_chars": 900,
+  "min_score": 6.0,
+  "drop_junk": true
+}