fix(plugin): preserve structure on PreCompact + capture compact summary

Two changes to the compaction flow:

PreCompact: agent stores its summary with infer=False
  on_pre_compact.sh now instructs the agent to call add_memory with
  infer=False. The agent has full conversation context and writes a
  structured summary; without infer=False the platform runs a second
  LLM extraction pass over that summary, losing structure and
  producing fragmented facts. With it, the structured text is
  preserved verbatim. Same guidance added to mem0-mcp/SKILL.md so
  the agent applies it to other already-extracted writes (decisions,
  anti-patterns, conventions).

SessionStart-compact: capture the platform-generated summary
  PreCompact fires BEFORE Claude Code generates its compact summary,
  so the summary is unreachable from that hook. SessionStart-compact
  fires AFTER, so a new background helper (capture_compact_summary.py)
  reads the transcript at that point, finds the isCompactSummary=true
  entry, and stores its content as a separate memory tagged
  metadata.type=compact_summary. The post-compaction bootstrap text
  now tells the agent to layer its recall queries across
  session_state (its own pre-compact summary), compact_summary (the
  platform's condensed view), and topical types like decision /
  anti_pattern.

Net effect: three complementary memory types from one session
boundary, none of which used to be filterable or even captured
correctly.

mem0-plugin/skills/mem0-mcp/SKILL.md

@@ -99,6 +99,21 @@ Extract key learnings and store them using the `add_memory` tool:
 
 Memories can be as detailed as needed -- include full context, reasoning, code snippets, file paths, and examples. Longer, searchable memories are more valuable than vague one-liners.
 
+### Use `infer=False` for already-structured content
+
+When you've done the extraction work yourself — pre-compaction summaries, decisions, anti-patterns, conventions you've explicitly identified — pass `infer=False` so the platform stores your text verbatim instead of running a second extraction pass over it.
+
+```python
+add_memory(
+    messages=[{"role": "user", "content": "<your structured fact>"}],
+    user_id="<active user_id>",
+    metadata={"type": "decision"},
+    infer=False,
+)
+```
+
+Stick to one mode per distinct piece of content — don't mix `infer=True` (default) and `infer=False` for the same fact, you'll get duplicates. Default (`infer=True`) is right for raw conversational signal you want extracted; `infer=False` is right for pre-extracted structure.
+
 ## Before losing context
 
 If context is about to be compacted or the session is ending, store a comprehensive session summary: