Fix title parser and feedback component routing in schematics

Title parser: always use line 0 as title per SPICE convention instead
of skipping * comment lines (which consumed component definitions).

Layout engine: pass growing placed set to path tracing to prevent
feedback components (e.g. col→base cap) from being double-traced.
Route feedback paths from the output terminal horizontally right
instead of defaulting to "down" through the transistor body.

backend/src/spicebook/engine/schematic.py

@@ -122,15 +122,12 @@ def parse_netlist(text: str) -> ParsedNetlist:
     if not lines:
         return ParsedNetlist()
 
-    # First non-blank, non-directive line is the title (SPICE convention)
-    title = ""
-    first_content = 0
-    for i, line in enumerate(lines):
-        stripped = line.strip()
-        if stripped and not stripped.startswith(("*", ";", ".")):
-            title = stripped
-            first_content = i + 1
-            break
+    # SPICE convention: line 1 is always the title, regardless of content.
+    # Strip leading * or ; (common title prefix) for display purposes.
+    title = lines[0].strip()
+    if title.startswith(("*", ";")):
+        title = title[1:].strip()
+    first_content = 1
 
     # Pre-scan for .model statements to determine BJT/MOSFET polarity
     models: dict[str, str] = {}
@@ -503,7 +500,7 @@ def _plan_active_layout(
     placed = set(exclude)
 
     for tname, tnode in terminals.items():
-        tpaths = _trace_paths_from_terminal(tnode, tname, node_map, exclude)
+        tpaths = _trace_paths_from_terminal(tnode, tname, node_map, placed)
         paths[tname] = tpaths
         for p in tpaths:
             for c in p.components:
@@ -557,6 +554,11 @@ def _path_style(
         return "output"
     if path.end_type == "ground":
         return "down"
+    # Feedback / coupling paths ending at non-supply, non-ground nodes
+    # (e.g., collector→base feedback cap): route horizontally outward
+    # from the output terminal to avoid crossing through the body.
+    if term_name == output_term:
+        return "output"
     return "down"