feat(slack): add per-channel natural thread routing

[KCAYAAI]

Summary

Add slack.natural_thread_channels, a per-channel routing control for Slack workspaces that need mention-to-start, then natural thread replies in selected collaboration channels while keeping strict mention-only behavior everywhere else.

The option is empty by default and does not change existing installations.

Problem

The existing controls operate at the wrong scope for this workflow:

• strict_mention: false or thread_require_mention: false weakens thread gating across the whole Slack profile.
• free_response_channels also admits unmentioned top-level messages, so it cannot express “mention to start.”
• require_mention_channels tightens channels; it does not provide a channel-scoped exception to strict thread gating.

Multi-agent and shared-workspace deployments therefore have to choose between repetitive mentions in designated collaboration threads or profile-wide auto-follow behavior in unrelated channels.

Behavior

slack:
  require_mention: true
  strict_mention: true
  thread_require_mention: true
  natural_thread_channels:
    - C0123456789

For a listed channel:

1. An unmentioned top-level message is ignored, including when the profile is otherwise free-response.
2. An explicit bot mention starts engagement and records the thread.
3. Plain human replies in that engaged thread can continue naturally.
4. Existing bot-message, active-session, parent-mention, authorization, allowed-channel, and bot-loop protections continue to apply.

For every unlisted channel, the existing strict_mention and thread_require_mention behavior is unchanged.

Changes

• Bridge natural_thread_channels from config.yaml into the Slack adapter.
• Compute strict/thread mention policy per channel instead of weakening profile-wide booleans.
• Force the top-level mention gate for listed channels, even if they overlap a free-response setting.
• Add routing tests for YAML/list/CSV parsing, mention-to-start, natural thread follow-up, free-response precedence, and strict unlisted channels.
• Document the option and its relationship to the existing mention controls.

Related mention-routing PRs

• fix(slack): trust adapter routing after stripping self mention #70238 — stripped-self-mention routing. Its adapter change is in a separate preprocessing area and this PR does not replace it.
• fix(slack): accept strict-mention clarification text replies #71044 — strict-mention clarification replies. It touches the same Slack handler and YAML bridge, so whichever PR merges second may need a narrow rebase; the behaviors and regression suites remain separate.

Validation

• Focused routing suite: 97 passed
• All 32 Slack test files: 893 passed
• Ruff: passed
• git diff --check: passed
• Docusaurus production build: passed for English and zh-Hans

The Slack-wide suite retains 37 existing non-failing asynchronous mock cleanup warnings.

[teknium1]

Thanks for the focused Slack routing change. Current main still globally rejects unmentioned messages at plugins/platforms/slack/adapter.py:5693-5716 when strict_mention or thread_require_mention applies, so the requested per-channel natural-thread behavior remains needed. I found no correctness defect in the proposed routing path.

Suggested changes

• Add natural_thread_channels: "" to the Slack defaults in hermes_cli/config_defaults.py:1783-1797, alongside the existing Slack routing controls. The adapter has a safe empty fallback, but the canonical defaults should enumerate the documented configuration surface.

Automated hermes-sweeper review.