Auto-strip Nevo preambles and separate style analysis per appeal subtype

- Add strip_nevo_preamble() to extractor.py — auto-removes Nevo database
  headers (bibliography, legislation, mini-ratio) during training upload
- Add appeal_subtype column to style_patterns table — patterns are now
  stored per subtype instead of globally mixed
- Update clear_style_patterns() to support subtype-scoped deletion
- Pass appeal_subtype through analyze_corpus → store → upsert pipeline

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

mcp-server/src/legal_mcp/services/style_analyzer.py

@@ -134,8 +134,8 @@ async def analyze_corpus(appeal_subtype: str = "") -> dict:
     if not rows:
         return {"error": "אין החלטות בקורפוס. העלה החלטות קודמות תחילה."}
 
-    # Clear old patterns before re-analysis
-    await db.clear_style_patterns()
+    # Clear old patterns for this subtype (or all if unfiltered)
+    await db.clear_style_patterns(appeal_subtype)
 
     # Calculate token budget
     total_chars = sum(len(row["full_text"]) for row in rows)
@@ -147,12 +147,12 @@ async def analyze_corpus(appeal_subtype: str = "") -> dict:
     )
 
     if estimated_tokens < MAX_INPUT_TOKENS:
-        return await _analyze_single_pass(rows)
+        return await _analyze_single_pass(rows, appeal_subtype)
     else:
-        return await _analyze_multi_pass(rows)
+        return await _analyze_multi_pass(rows, appeal_subtype)
 
 
-async def _analyze_single_pass(rows) -> dict:
+async def _analyze_single_pass(rows, appeal_subtype: str = "") -> dict:
     """Send all decisions in a single API call."""
     decisions_text = ""
     for row in rows:
@@ -164,10 +164,10 @@ async def _analyze_single_pass(rows) -> dict:
         timeout=claude_session.LONG_TIMEOUT,
     )
 
-    return await _parse_and_store_patterns(raw, len(rows))
+    return await _parse_and_store_patterns(raw, len(rows), appeal_subtype)
 
 
-async def _analyze_multi_pass(rows) -> dict:
+async def _analyze_multi_pass(rows, appeal_subtype: str = "") -> dict:
     """Analyze each decision individually, then synthesize patterns."""
     all_patterns = []
 
@@ -197,7 +197,7 @@ async def _analyze_multi_pass(rows) -> dict:
         timeout=claude_session.LONG_TIMEOUT,
     )
 
-    return await _parse_and_store_patterns(raw, len(rows))
+    return await _parse_and_store_patterns(raw, len(rows), appeal_subtype)
 
 
 def _extract_json(response_text: str) -> list | None:
@@ -248,14 +248,16 @@ def _extract_json(response_text: str) -> list | None:
     return None
 
 
-async def _parse_and_store_patterns(response_text: str, num_decisions: int) -> dict:
+async def _parse_and_store_patterns(
+    response_text: str, num_decisions: int, appeal_subtype: str = "",
+) -> dict:
     """Parse Claude's response and store patterns in the database."""
     patterns = _extract_json(response_text)
 
     if patterns is None:
         return {"error": "Could not parse analysis results", "raw": response_text}
 
-    # Store patterns
+    # Store patterns tagged by appeal_subtype
     count = 0
     for pattern in patterns:
         await db.upsert_style_pattern(
@@ -263,11 +265,13 @@ async def _parse_and_store_patterns(response_text: str, num_decisions: int) -> d
             pattern_text=pattern.get("text", ""),
             context=pattern.get("context", ""),
             examples=[pattern.get("example", "")],
+            appeal_subtype=appeal_subtype,
         )
         count += 1
 
     return {
         "patterns_found": count,
         "decisions_analyzed": num_decisions,
+        "appeal_subtype": appeal_subtype or "all",
         "pattern_types": list({p.get("type") for p in patterns}),
     }