feat(discord): fetch thread history when bot is first mentioned mid-thread

[AgentWrapper]

Summary

When the Hermes bot is @mentioned in an existing Discord thread for the first time, it currently starts a blank session with no awareness of the thread's prior conversation. This PR ports the Slack adapter's _fetch_thread_context() pattern to Discord, so the agent receives the thread history as context on first mention.

Closes #6708

Changes

Three additions to gateway/platforms/discord.py:

1. _has_active_session_for_thread() — checks the session store via build_session_key() to determine if a session already exists for this thread. Respects thread_sessions_per_user and group_sessions_per_user config settings. Mirrors the Slack adapter's implementation from feat(gateway): approval buttons for Slack & Telegram + Slack thread context #5890.

2. _fetch_thread_context() — fetches up to 30 prior messages from the Discord thread via channel.history(oldest_first=True), formats them as:

   [Thread context — previous messages in this thread:]
   User1: message text
   User2: message text
   [End of thread context]
   

   Skips the bot's own messages and the current trigger message. Strips bot @mentions from context messages.

3. Injection in _handle_message() — when the message is in an existing thread (not auto-created), and there's no prior session for it, calls _fetch_thread_context() and prepends the result to event_text. Follow-up messages use the normal session transcript.

Design decisions

• Only fires for existing threads, not auto-created ones — when auto_thread creates a brand-new thread, there's no history to fetch, so we skip the API call.
• Uses build_session_key() as single source of truth — avoids the manual key construction bug that was fixed in the Slack adapter (fix/slack-thread-session-key-mismatch  #5833), where thread_sessions_per_user was ignored.
• Matches Slack formatting exactly — same [Thread context] ... [End of thread context] envelope so the agent prompt is consistent across platforms.
• Limit of 30 messages — same as Slack, keeps the context injection bounded.

Testing

• All 154 existing Discord tests pass
• Syntax verification passes
• No changes to existing lines (113 insertions, 0 modifications)

[Tranquil-Flow]

Heads up — I've opened #7466 which takes a similar approach but generalizes it into a base class hook (BasePlatformAdapter.fetch_thread_context()). The Discord override in that PR follows the same pattern as your implementation here (using channel.history(), skipping bot messages, stripping mentions, same 30-message limit).

The motivation was to avoid each platform adapter reimplementing the same session-check + context-fetch + prepend pattern independently. The base class also handles command safety (skipping /reset etc.) and session expiry checks.

Happy to coordinate if you'd prefer a different approach or want to merge aspects of both.

[alt-glitch]

Related to #7466 (unified thread context hook in BasePlatformAdapter) which addresses the same issue #6708 with a more general approach.

[teknium1]

Thanks for the focused Discord implementation. This is an automated hermes-sweeper review; current main already provides the requested behavior.

• e84fe483bc958ef2ce11463d10ee57bdc2ccc5fb added Discord channel-history backfill and is an ancestor of current main.
• plugins/platforms/discord/adapter.py:6532 calls _fetch_channel_context() for existing threads and skips only newly auto-created threads.
• plugins/platforms/discord/adapter.py:4995 fetches bounded pre-trigger history and formats it as agent context; tests/gateway/test_discord_free_response.py:1358 covers thread backfill dispatch.
• The adapter has since moved from this PR's gateway/platforms/discord.py target to plugins/platforms/discord/adapter.py.

The related discussion of #7466 was considered; this close is based on the independently verified behavior already present on current main.