Fix refrag GLM mode, add GLM support to CTX, ctx based Claude Code hook to enhance user prompts

[voarsh2]

This PR:

• Adds GLM support to ReFrag (without LLM sidecar)
• Allows CTX to use GLM to rewrite prompts, extra relevance gating flags (which can just return original query/prompt)
  -- ctx config can set collection and gating/relevance knobs, with example provided)
• Adds a Claude Code hook script and hook example, to rewrite the users prompt and give Claude Code the enhanced input.
  -- hook can be bypassed with adding "<" to msg, or "noctx:" - planning to add "only enhance IF (a condition is met) - right now it's enhance unless opt-out

Sample CTX prompt rewrite in Claude Code:

echo '{"user_message":"Describe how the quest progression system in Pirates Online Classic is implemented. How are quest states represented, how are objectives registered and updated, and where do we validate that a quest step is complete before advancing to the next stage?"}' | ./ctx-hook-simple.sh
{
  "user_message": "Analyze the quest state representation in QuestStepIndicator by examining the specific methods and attributes defined between lines 33-43 in /work/pirates-online-classic/pirates/quest/QuestStepIndicator.py. How do these components track quest progress, and what data structures are used to maintain quest state? Compare this with the quest registration mechanisms in QuestManagerAI (lines 454-483), focusing on how objectives are registered and updated. How does QuestManagerAI establish communication with DistributedBuriedTreasureAI (lines 30-49) to synchronize quest progress, and what specific message passing or event handling patterns are employed?\n\nInvestigate the quest step validation process by analyzing the conditional checks in DistributedBuriedTreasureAI (lines 30-49) and DistributedSearchableContainerAI (lines 23-36). What specific criteria determine quest step completion in buried treasure quests, and how are these conditions evaluated? How does QuestManagerAI (lines 454-483) process validation results from DistributedBuriedTreasureAI to advance quest stages, and what error handling or edge cases are considered? What role does DistributedSearchableContainerAI play in this?"
}```

[voarsh2]

@m1rl0k lol I wasn't done yet - I forgot to provide the example Claude hook docs and finish testing