fix(unreads): fall back when client.counts is team_is_restricted

[arnaud-gp]

Problem

On some Enterprise Grids client.counts is blocked org-wide with team_is_restricted. That is not a token-host mismatch: the same error is returned on both the org shell and the workspace host, with or without org_wide_aware. agent-slack unreads (and channel list --via-counts) then exit 1 even though client.dms, users.conversations, conversations.info, and conversations.history still work.

What Changed

• Keep client.counts as the primary unreads path. Fall back only on team_is_restricted.
• DM / MPIM fallback: client.dms with count 50 (most recently active conversations), then conversations.info (last_read) compared to the latest message ts. Call volume is 1 + ≤50 info calls and does not sweep the full IM list.
• Channel fallback: users.conversations (public/private, max 50, newest updated first when present), then conversations.info plus conversations.history.
• channel list --via-counts uses the same trigger and falls back to users.conversations.
• later list uses the same trigger on saved.list and falls back to search.messages with is:saved.
• README documents the fallback and the DM call bound.

Verification

• bun test — 412 pass, 0 fail. New test/unreads.test.ts covers the team_is_restricted trigger, the counts-only DM bound, and the primary client.counts path. Channel and Later fallbacks are covered in their existing test files.
• bun run typecheck — clean.

Risks

• Fallback unreads omit thread-unread counts (those only come from client.counts).
• Channel scan is capped at 50 joined conversations, so a quieter unread channel outside that window can be missed.
• Later fallback via is:saved cannot distinguish archived/completed states; those filters return empty items with a total count only.

Made with Cursor