fix(slack): avoid null table column settings

[tianma-if]

Summary

• Replace invalid null Slack table column_settings entries with schema-valid objects and omit trailing default columns.
• If Slack rejects a rendered chat.postMessage payload with invalid_blocks, retry once as plain text.
• Apply the same fallback to finalized streaming chat.update calls, using blocks=[] so an existing block layout is cleared.
• Preserve workspace routing by resolving the metadata-selected client once and reusing it for both attempts.

Tests

• Table settings contain only objects and omit trailing defaults.
• Post fallback retries without blocks on the same workspace client.
• Final edit fallback clears blocks on the same workspace client.
• .venv/bin/python -m pytest tests/gateway/test_slack_block_kit.py tests/gateway/test_slack_block_kit_adapter.py tests/gateway/test_slack.py -q (277 passed)
• .venv/bin/python -m ruff check plugins/platforms/slack/adapter.py plugins/platforms/slack/block_kit.py tests/gateway/test_slack_block_kit.py tests/gateway/test_slack_block_kit_adapter.py
• git diff --check

[alt-glitch]

This was generated by AI during triage.

Competes with open PR #56618, which also fixes #56615 (null column_settings → invalid_blocks) by hardening the merged #56102 Block Kit path. Same goal, independent implementations — flagging for a maintainer to pick one; not marking either as a duplicate. Related: #56615 (issue), #56102 (the merged rich-blocks feature this hardens).

[teknium1]

Thanks for targeting the live rich-table failure; current main still emits None placeholders in plugins/platforms/slack/block_kit.py:288-300.

Problems

• The new client = self._get_client(chat_id) in this PR's plugins/platforms/slack/adapter.py:1407 drops current workspace routing. Main's live send path uses _get_client(chat_id, team_id=self._metadata_team_id(metadata)) at plugins/platforms/slack/adapter.py:1447-1449; preserve that on both retry attempts and test it with metadata.
• The no-block recovery is only added to send(). Finalized streaming edits also render blocks and call chat_update in plugins/platforms/slack/adapter.py:1541-1547, so an invalid_blocks response there still fails. Add a matching fallback that sends blocks=[] to clear a previous layout.

Suggested changes

• Salvage the retry into the current workspace-aware send call and assert both client selections.
• Add the finalized-edit fallback and regression coverage.

This is an automated hermes-sweeper review.

[teknium1]

Current main routes outbound sends through _get_client(chat_id, team_id=self._metadata_team_id(metadata)) (plugins/platforms/slack/adapter.py:1447-1449). Retain that metadata on the initial post and retry; this unqualified lookup can route a multi-workspace response through the primary client.

[tishan-jayasekera]

Verified at the current head (aff6d4835): the metadata-selected client is resolved with team_id=self._metadata_team_id(metadata) and reused for both the initial request and fallback. The adapter regression test asserts _get_client.assert_called_once_with("C1", team_id="T_OTHER"). The focused Slack suite passes 277/277. This concern is addressed; leaving thread resolution to a maintainer.

[tishan-jayasekera]

Independent validation evidence for this PR:

• reproduced invalid_blocks / must provide an object when a mixed-alignment table emitted a null column_settings placeholder
• confirmed an object-only placeholder payload posts successfully in a private DM canary
• checked out this PR's exact head (aff6d4835) and ran scripts/run_tests.sh tests/gateway/test_slack_block_kit.py tests/gateway/test_slack_block_kit_adapter.py tests/gateway/test_slack.py -q: 277 passed, 0 failed
• git diff --check origin/main...HEAD is clean
• the earlier sweeper concerns are addressed at the current head: workspace-specific client routing is preserved and finalized chat.update calls have the same fallback

This renderer correction plus the plain-text retry is more complete than my narrower duplicate #68434, which I closed in favor of this canonical PR.

@teknium1, could you please take the maintainer merge look when convenient? The PR is currently mergeable with all required checks green.

[teknium1]

Closing as superseded by #69317 (merged): same column_settings fix as #56618, submitted later; #56618 was credited as earliest.

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