fix(slack): chunk section-block content and surface postMessage errors (#18)

[idl3]

Summary

When an agents sends a ~2KB structured report via send_slash_confirm, Slack silently rejected it because the section block text field exceeded the 3000-char-per-element API limit — producing two empty-looking messages at the same timestamp. The top-level text field was fine (well under 40k), but the block-level limit is separate and was never enforced.

Root cause: send_slash_confirm used a hard truncation at 2900 chars (message[:2900] + "..."), which silently drops content rather than chunking it. More critically, no call site checked the Slack API response for error details, so the rejection was invisible in logs.

This PR makes the send path section-block-aware and adds error surfacing across all chat_postMessage call sites:

• MAX_SECTION_TEXT_LENGTH = 2900 constant added near MAX_MESSAGE_LENGTH (Slack's 3000-char hard limit minus 100-char margin).
• send_slash_confirm rewritten: uses BasePlatformAdapter.truncate_message to split the message body at the section budget (preserving code-block boundaries). The first chunk goes in the button block; overflow posts as threaded plain-text replies with [i/N continued] markers — no content is ever silently dropped.
• _safe_post_message helper wraps every chat_postMessage call. On SlackApiError it logs error_code, text_len, blocks_count, and longest section-block text length at WARNING level before re-raising. Wired into send(), send_exec_approval, send_slash_confirm, and the button-action follow-up path.
• 10 new tests in tests/gateway/test_slack_chunking.py covering: large/small send() chunking, section-block size ceiling, continuation threading, full content preservation, and _safe_post_message error-code logging. All 285 existing Slack tests continue to pass; ruff clean.

new-pattern: yes
what: send_slash_confirm now chunks at section-block level and routes all chat_postMessage calls through a logging wrapper
visualizable: no
why: the change is a linear call-site refactor with no branching flow or state machine; a diagram would add no information over the prose description
decision: skip

Test plan

• uv run --extra dev python -m pytest tests/gateway/test_slack_chunking.py -v — 10/10 pass
• uv run --extra dev python -m pytest tests/gateway/test_slack.py tests/gateway/test_slack_approval_buttons.py tests/gateway/test_slack_mention.py tests/gateway/test_slack_channel_skills.py -v — 275/275 pass
• uv run --extra dev ruff check gateway/platforms/slack.py tests/gateway/test_slack_chunking.py — clean
• Reviewer: send a >3000-char report via /hermes slash command in a real workspace and confirm no empty-message pair appears

Closes #18

[teknium1]

Thanks for addressing the Slack block-limit failure. Current main already prevents the rejected-block symptom, but this PR needs a targeted port and one boundary fix.

Problems

• gateway/platforms/slack.py was moved to plugins/platforms/slack/adapter.py by 1a38066054752d601b71fc655a3ed6bf4228e2da, so the submitted patch no longer targets the active adapter.
• The new title_str = title or "Confirm" at gateway/platforms/slack.py:2390 is unbounded. A sufficiently long title makes header exceed Slack's 3000-character section limit before body content is considered. Current main caps this input at plugins/platforms/slack/adapter.py:3348.

Suggested changes

• Port the continuation/error-detail work to plugins/platforms/slack/adapter.py while retaining the current 150-character title cap and rendered-size budget.
• Add an overlong-title regression test; current main already has the relevant section-limit guard in the active adapter.

Automated hermes-sweeper review.

[teknium1]

Please retain a title cap here (current main uses [:150]). Without one, a long title makes header exceed Slack's 3000-character section limit before any body chunk is added.

[teknium1]

Closing as superseded by #69317 (merged): targeted the pre-plugin-migration file; section chunking already exists on main (_split_text) and error surfacing landed via #56618's retry-fallback.

Thanks for digging into this — the consolidated fix stands on the cluster's collective analysis, and your work is credited in #69317's summary.