Implement D3-D7: legal NER, financial extraction, speaker ID, re-extraction, quality stats

D3: Legal NER — regex pattern extraction for US/UK case citations, statute
    references, court names, attorney/judge identification
D4: Financial extraction — currency amounts, IBAN/SWIFT/BIC, account numbers,
    wire transfer references, date-amount pairing
D5: Speaker identification — deposition Q/A pattern detection, speaker turn
    tracking, word count per speaker, cross-talk detection
D6: Batch re-extraction — stage diff planning, Cap'n Proto result merging,
    selective stage re-run with mask comparison
D7: Quality dashboard — per-stage success/failure counts, OCR confidence
    histograms, processing time percentiles, dedup hit rates, cross-locale
    merge support

5 new Zig modules (3,399 lines), 1 modified FFI dispatcher.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: hyperpolymath

ffi/zig/src/docudactyl_ffi.zig

@@ -24,6 +24,11 @@ const conduit = @import("conduit.zig");
 const gpu_ocr = @import("gpu_ocr.zig");
 const hw_crypto = @import("hw_crypto.zig");
 const ml_inference = @import("ml_inference.zig");
+const legal_ner = @import("legal_ner.zig");
+const financial_extract = @import("financial_extract.zig");
+const speaker_id = @import("speaker_id.zig");
+const reextract = @import("reextract.zig");
+const quality_stats = @import("quality_stats.zig");
 
 // Ensure submodule exports are included in the shared library
 comptime {
@@ -34,6 +39,11 @@ comptime {
     _ = gpu_ocr;
     _ = hw_crypto;
     _ = ml_inference;
+    _ = legal_ner;
+    _ = financial_extract;
+    _ = speaker_id;
+    _ = reextract;
+    _ = quality_stats;
 }
 
 const c = @cImport({