[4.41/SL4b] Slack inbound mrkdwn normalization, channel.post thread ids, Socket Mode retries (remainder of #209)

[patrick-chinchill]

Summary

Remainder of #209, split per its Metadata (the full port came to about 2k changed lines; the split threshold is 1.5k). #209's first PR ports part (a): self-mention decoding, content-based is_mention, and bot author ids / email / is_system. This issue is part (b): inbound mrkdwn normalization, channel.post thread ids, and Socket Mode retries.

Starting point: branch wip/4.41-sl4-full (commit c0788056) holds the full #209 port, with full validation green on its base: 6710 passed, strict fidelity 733/733, pyrefly 0 errors. Part (b) is that branch's diff minus what the #209 part-(a) PR merged. Rebase or cherry-pick the (b) pieces listed below once that PR has landed, re-run the validation, and send the result through astra review. Nothing on that branch has been reviewed yet.

Upstream changes

• e71bfead fix(slack): preserve first line of incoming code blocks (#843), chat@4.39.0
• 44423bdc fix(slack): convert special mentions to readable text (#960), chat@4.41.1
• c3118279 fix(slack): preserve channel id and normalize frequently hallucinated link format (#756), chat@4.37.0
• 92530dd3 fix(slack): return replyable thread ID from channel posts (#720), chat@4.35.0
• 0b63791b fix(slack): process Socket Mode retry envelopes instead of dropping them (#667), chat@4.33.0 (Slack half)

Scope (all already on wip/4.41-sl4-full)

• adapters/slack/format.py: port convertSpecialMentions and convertMrkdwnWithCodeFences and their helpers (findAngleTokenEnd, findInlineCodeEnd, isOnBlockquoteLine, escapeLeadingBlockMarker), plus the #name (C…) rewrite and the inverted-link swap. The scanners are linear: a failed <…> scan is memoized, the newline search is bounded, and the blockquote check matches in place.
• SlackFormatConverter.to_ast becomes parse_markdown(slack_mrkdwn_to_markdown(text)). This fixes the missing unescape, so < now renders as <. Drop the Python-only regex extract_plain_text override; upstream markdown.ts has none.
• Both parse paths: formatted = to_ast(text), then text = ast_to_plain_text(formatted) (upstream toPlainText(formatted)). [4.41/SL5] Slack inbound: pasted tables and alert attachments in message content #210 later swaps formatted for the assembled content.
• post_channel_message: return slack:{channel}:{ts} when raw["ts"] is a str.
• Socket Mode: stop dropping retry envelopes. Log "Processing socket mode retry" at info with retry_attempt, retry_reason and type, then route the event.
• Tests:
  • tests/test_slack_format_primitives.py: every it that e71bfead (10), 44423bdc (7, including the it.each) and c3118279 (3) added, plus an adversarial 50k-character case.
  • tests/test_slack_format.py: the 6 e71bfead markdown.test.ts cases.
  • tests/test_slack_socket_mode.py: "processes retries like first deliveries (dedupe drops true duplicates)", replacing test_retry_attempt_is_skipped_but_acked.
  • Slack parse tests: "converts special mentions to readable text", "preserves special mention tokens in an inbound inline code span" and "preserves special mention tokens in an inbound code block".
  • Post-channel tests: "posts to channel without thread context", plus a Python case with no string ts.
• docs/UPSTREAM_SYNC.md and the CHANGELOG Breaking/consumer-visible bullets for: special mentions and #general (C123), the kept first code line, message.text as plain text of formatted (list markers, #, > and backticks dropped), channel.post() ids, and processed socket retries.

Known gaps found while porting (not Slack-specific; record them or file separately)

• The shared parse_markdown has no multi-backtick code spans and keeps a paragraph's leading space. Two markdown.test.ts assertions were adapted ("keeps a quoted fence inside the blockquote", "keeps trailing text after a code block as a paragraph").
• The shared parse_markdown is quadratic on some inputs: 50k characters of `x or *a take 25–37 s. This predates the port, but every inbound Slack message goes through to_ast.

Notes

• Consumer impact: high. The shape of message.text changes. A live Slack-loop check is pending.
• Links: [4.41/SL4] Slack inbound normalization: self-mention decode, content is_mention, mrkdwn fixes, bot author ids, socket retries #209 (parent), [4.41] Tracking: sync upstream chat@4.31.0 → chat@4.41.1 (0.4.41 wave) #184 (wave tracking). Blocks [4.41/SL5] Slack inbound: pasted tables and alert attachments in message content #210, together with [4.41/SL4] Slack inbound normalization: self-mention decode, content is_mention, mrkdwn fixes, bot author ids, socket retries #209 part (a).

[patrick-chinchill]

Coordination note from #291 (#210): #291 no longer needs to wait for this issue. Until #283 lands, SlackAdapter._assemble_content takes the body's share of message.text from the pre-#210 extract_plain_text and appends the tables' and attachments' ast_to_plain_text. If #291 merges first, the "Both parse paths" item here becomes: replace that with ast_to_plain_text(formatted), then update the four interim assertions listed in docs/UPSTREAM_SYNC.md ("Temporary divergence until #283: the body's plain text").