scripts: use capnp store instead of reading markdown directly

Add store_helpers.py with shared helpers that call poc-memory commands
(list-keys, render, journal-tail) instead of globbing ~/.claude/memory/*.md
and parsing section headers.

All 9 Python scripts updated: get_semantic_keys(), get_topic_file_index(),
get_recent_journal(), parse_journal_entries(), read_journal_range(),
collect_topic_stems(), and file preview rendering now go through the store.

This completes the clean switch — no script reads archived markdown files.

scripts/bulk-categorize.py

@@ -68,20 +68,22 @@ def get_unique_files(keys: list[str]) -> list[str]:
 
 def build_prompt(files: list[str]) -> str:
     """Build categorization prompt."""
-    # Read first few lines of each file for context
+    # Read file previews from the store
     file_previews = []
     for f in files:
-        path = MEMORY_DIR / f
-        if not path.exists():
-            # Try episodic
-            path = MEMORY_DIR / "episodic" / f
-        if path.exists():
-            content = path.read_text()
-            # First 5 lines or 300 chars
-            preview = '\n'.join(content.split('\n')[:5])[:300]
-            file_previews.append(f"  {f}: {preview.replace(chr(10), ' | ')}")
-        else:
-            file_previews.append(f"  {f}: (file not found)")
+        try:
+            r = subprocess.run(
+                ["poc-memory", "render", f],
+                capture_output=True, text=True, timeout=10
+            )
+            content = r.stdout.strip()
+            if content:
+                preview = '\n'.join(content.split('\n')[:5])[:300]
+                file_previews.append(f"  {f}: {preview.replace(chr(10), ' | ')}")
+            else:
+                file_previews.append(f"  {f}: (no content)")
+        except Exception:
+            file_previews.append(f"  {f}: (render failed)")
 
     previews_text = '\n'.join(file_previews)