fix(feishu): preserve markdown around tables

[lazydao]

Summary

Fixes Feishu/Lark markdown rendering when an outgoing message contains Markdown tables.

Previously, any content that looked like a table forced the entire outgoing message to msg_type=text. That kept the table visible but exposed headings, emphasis, inline code, links, and other Markdown markers as raw text.

Details

• Keeps table-containing messages on Feishu post payloads.
• Emits table blocks as plain text rows because Feishu post md rows do not render tables reliably.
• Preserves surrounding prose as md rows.
• Leaves fenced code blocks untouched.
• Adds regression coverage for mixed Markdown/table segmentation and the outbound send path.

This is intentionally narrower than the Card v2 table approach in #12114: it preserves existing post-message behavior and client compatibility without switching the whole response to an interactive card.

Validation

• scripts/run_tests.sh tests/gateway/test_feishu.py -q
• Result: 212 passed
• git diff --check

[alt-glitch]

This was generated by AI during triage.

Competing approach in the saturated Feishu markdown-table cluster (canonical open #12114; tracked by #27469). This PR keeps table-containing messages on post payloads and splits the table block into plain-text rows while leaving surrounding prose as md rows — distinct from the card-table rendering PRs (#12114, #22272, #35906). Not a duplicate; a maintainer should pick the canonical fix for this cluster.

[teknium1]

Thanks for the focused Feishu fix. The premise remains present on current main: _build_outbound_payload() sends any detected table as text at plugins/platforms/feishu/adapter.py:4524-4533, which bypasses the adjacent markdown-post path. This patch's table-to-text-row segmentation (bd97a0176f21, plugins/platforms/feishu/adapter.py:617-624) preserves surrounding md rows and intentionally skips fenced code blocks.

The PR's added coverage exercises mixed markdown/table segmentation and verifies the send path selects post. The related discussion identifies Card v2 table rendering in #12114 as a distinct competing approach, so choosing a canonical rendering strategy remains a maintainer decision rather than a correctness concern with this patch.

Automated hermes-sweeper review.

[teknium1]

Closing — fixed on main by #68121. Your approach (table blocks as plain-text rows inside an otherwise markdown-rendering post) solved the real complaint — one table no longer destroys the whole message's formatting — but Feishu's server-side fix made the special-casing unnecessary: post/md now renders GFM tables natively, so the whole message renders. Thanks!