[codex] illustrate Slack stream payload budgeting

[Zygimantass]

Draft PR for illustration only. This is meant to show the adapter-side changes we made downstream to make Slack native streaming work reliably for Centaur, not as a polished upstream API proposal.

What this demonstrates:

• rotate Slack native stream segments before the stream age/size limits are hit
• estimate combined markdown plus structured task/plan payload size, instead of only markdown length
• split large task updates into Slack-sized chunks
• annotate stream delivery failures with slackStreamMessageId and slackAnswerLost so an app-level durable fallback can replace the failed stream instead of posting a duplicate
• tests for markdown rotation, structured task payload rotation, and msg_too_long annotation

Context from our incident:

Slack returned msg_too_long on chat.stopStream even though the final answer text was short. The oversized payload came from accumulated structured plan/task block JSON. In Centaur, the replacement fallback remains app-side; this PR only illustrates the SDK adapter behavior needed to support that.

Validation:

• pnpm exec ultracite check packages/adapter-slack/src/index.ts packages/adapter-slack/src/index.test.ts
• pnpm --filter @chat-adapter/slack typecheck
• pnpm --filter @chat-adapter/slack test -- index.test.ts
• pnpm --filter @chat-adapter/slack build

[vercel]

@Zygimantass is attempting to deploy a commit to the Vercel Team on Vercel.

A member of the Team first needs to authorize it.