NS_GSM Full Corpus Ingestion v0.2

The first time real corpus documents are actually brought into the package (46 markdown files under corpus/, spanning the C, RFP, MORP, X72, DCRP, FCBP, and PAM lines). Adds manifest-driven source discovery, deduplicated identity, deterministic markdown extraction, series profiling, a persistent candidate layer, resumable ingestion checkpoints, and cross-series comparison proposals. 46 documents → 1,460 candidate extraction records (the report stresses, verbatim, that these are extraction records, not a count of independent theorems) and 219 cross-series exact-statement proposals, all left at NEEDS_REVIEW, with 0 automatic promotions. Per-series coverage status: C (PARTIAL, 2 documents), RFP (COMPLETE_SERIES Cycle I, 15 documents), MORP (COMPLETE_SERIES Cycle VII, 7 documents), X72 (SELECTED_MILESTONES, 4 documents), DCRP (SELECTED_MILESTONES, 10 documents), FCBP (COMPLETE_SERIES Cycle VI, 7 documents), PAM (COMPLETE_INDEX_ARTIFACT, 1 document). States explicitly that this version introduces no DCRP106 research theorem.