refactor: split engram.py (793→425 lines) into 4 modules

Extract HTTP client (engram_http.py), LLM client (engram_llm.py),
and utility functions (engram_utils.py) from monolithic engram.py.
Update all imports in source and tests.

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

Author: aeromomo

scripts/engram_auto.py

@@ -929,7 +929,7 @@ def _print_summary(
         if totals:
             # Show pending token counts per thread
             storage = EngramStorage(self.workspace)
-            from claw_compactor.engram import _count_messages_tokens
+            from claw_compactor.engram_utils import count_messages_tokens as _count_messages_tokens
             print("Thread pending tokens:")
             for tid in sorted(totals.keys()):
                 pending = storage.read_pending(tid)
@@ -967,7 +967,7 @@ def print_status(workspace: Path, engram_cfg: Dict) -> None:
         pending = storage.read_pending(tid)
         obs = storage.read_observations(tid)
         ref = storage.read_reflection(tid)
-        from claw_compactor.engram import _count_messages_tokens
+        from claw_compactor.engram_utils import count_messages_tokens as _count_messages_tokens
         pt = _count_messages_tokens(pending)
         ot = estimate_tokens(obs)
         rt = estimate_tokens(ref)

scripts/lib/cli.py

@@ -4,7 +4,7 @@
 import sys
 
 
-def main():
+def main() -> None:
     """Dispatch to mem_compress.py with rewritten arguments."""
     from pathlib import Path