feat: /close slash command to manually terminate a thread session

[qijie850]

Summary

Adds a Discord slash command /close that lets users manually terminate a coding session inside a thread.

Closes #40
Depends on #41 (per-thread working directories)

Changes

src/discord.rs:

• Register /close as a global slash command on ready
• Handle InteractionCreate events for the /close command
• Validates it's called inside a thread before closing

src/acp/pool.rs:

• Add close_session(thread_id) method that kills the ACP process and removes the per-thread working directory

Design Decision: Slash Command vs Text Matching

Per @thepagent's feedback on #40, this uses Discord slash commands (InteractionCreate event) instead of text matching to avoid shadowing agent-side /close commands:

Text message "/close"  ──> forwarded to ACP agent (agent's command)
Slash command /close   ──> handled by broker (Discord interaction)

Zero collision between broker commands and agent commands.

Behavior

• /close in a thread → kills session + cleans workdir → "\u2705 Session closed."
• /close with no active session → "\u2139\ufe0f No active session in this thread."
• /close outside a thread → "\u26a0\ufe0f Only works inside a thread."

Testing

Compiled and verified against serenity 0.12 on EC2 (Ubuntu 24.04). Slash command registration and interaction handling tested locally.

[thepagent]

Good work using Discord slash commands — exactly the approach we discussed in #40. A few items:

Must fix:

1. Overlapping diff with #41 — This PR includes the same pool.rs per-thread dir changes from #41. Please rebase onto #41's branch (or wait for #41 to merge, then rebase). Otherwise we'll get merge conflicts.

2. Carry over #41 review fixes — The pool.rs changes here have the same issues flagged in #41 review:

   • Path traversal hardening (thread_id validation)
   • shutdown() directory cleanup
   • sync exists() → use tokio::fs or just handle NotFound
   • Extract thread_dir() helper (now duplicated 3 times)

3. Make response ephemeral — /close response is visible to everyone. Add .ephemeral(true) so only the caller sees it:

CreateInteractionResponseMessage::new()
    .content(response_content)
    .ephemeral(true)

4. Global command registration on every restart — create_global_command in ready runs on every bot restart. Global commands have rate limits and take up to 1 hour to propagate. Consider:
   • Use guild commands instead (instant propagation, registered per-server)
   • Or gate with a one-time check

Looks good otherwise:

• Lock scope in close_session is well done (release write lock before async fs cleanup)
• Thread detection via thread_metadata.is_some() is correct
• Clean separation between slash command handling and text message flow

Suggested merge order: land #41 first → rebase this PR → address feedback → merge.

[qijie850]

Thanks for the review! Agreed on all points. Here's the plan:

Merge order: Will wait for #41 to land first, then rebase this PR on top — that automatically resolves #1 and #2 (the overlapping pool.rs changes + review fixes).

#3 — Ephemeral response: Will add .ephemeral(true) to the /close response. Makes sense since it's an admin action.

#4 — Guild commands: Good call on the rate limit issue. Will switch to guild commands (create_guild_command) for instant propagation and no 1-hour delay. We can gate it per-guild in ready.

Will push the fixes once #41 is merged and the rebase is clean!

[qijie850]

Rebased onto #41 (including all review fixes) and addressed all feedback:

Changes:

1. Rebased onto feat: per-thread isolated working directories #41 — no more overlapping pool.rs diff. This PR now only adds close_session() to pool and the slash command handler to discord.

2. Inherits feat: per-thread isolated working directories #41 fixes — path traversal hardening, shutdown() cleanup, thread_dir() helper, async-safe NotFound handling are all in the base.

3. Ephemeral response — added .ephemeral(true) so only the caller sees the confirmation.

4. Guild commands instead of global — registers /close per-guild in ready() using guild_id.create_command(). Instant propagation, no global rate limit concerns. Re-registration on restart is idempotent for guild commands.

5. Lock hygiene in close_session — releases write lock before async remove_dir_all, consistent with the pattern established in feat: per-thread isolated working directories #41.