fix: skip cross-workspace user collisions during bot sync

[steipete]

What Problem This Solves

Bot syncs of a multi-workspace archive abort at the trailing users snapshot: Slack's built-in slackbot has the literal ID USLACKBOT in every workspace, so the first workspace to record it owns the row and every later workspace's sync exits with WorkspaceCollisionError. Slack Connect can surface externally owned members the same way. Messages import fine; the run still ends nonzero and the workspace sync-state stamp is never written.

Why This Change Was Made

#130 fixed exactly this failure class for channels (Slack Connect shared channels) by demoting the collision to a skip-with-warning at the API sync boundary. The users snapshot was the missed sibling surface; the desktop import path (upsertDesktopUser) already tolerates user collisions the same way. The first workspace keeps the user row; all other upsert errors stay fatal.

User Impact

Multi-workspace archives sync cleanly again. Live repro: a Foundation-workspace bot sync on a database that also holds a desktop import of another workspace imported all messages but exited nonzero on USLACKBOT every run.

Evidence

• Regression test TestSyncSkipsUserOwnedByAnotherWorkspace: seeds USLACKBOT under another workspace, proves the sync completes, the original workspace keeps the row, other users still sync, and channel messages persist.
• GOWORK=off go test ./... green; make lint (vet + govulncheck) clean.

[steipete]

Closing as a duplicate subset of #131, which fixes the same users-snapshot collision (skipUserCollision in the same loop) and additionally covers the sibling surfaces this PR missed: message collisions in the socket-mode event handler, channel history, and thread sync, plus tail-repair resilience. The regression test here (TestSyncSkipsUserOwnedByAnotherWorkspace, seeding USLACKBOT under a second workspace) is available to cherry-pick if useful.