Revert "Merge branch 'tb-route-driven-chat-init-tier2' into 'master'"

Purpose of revert

Incident: gitlab-com/gl-infra/production#22875 (closed) (INC-13926, "68.9% of chat workflows are stopped", Sev-3).

Reverted !251389 (merged) on a time-correlation hypothesis: the MR shipped 3 Sep and made the client close the websocket (WORKHORSE_WEBSOCKET_CLOSE_1000) on thread select / new chat, which matched the dominant stop reason.

Outcome: hypothesis disproven. The revert reached production 2026-09-08 05:57 UTC; the stop rate did not move (68.9% in the 24h before → 62.8% in the 3h after). Root cause was upstream: Anthropic rate limiting on claude-fable-5-1 (8M input tokens/min) during a Fable usage spike, compounded by duo-workflow-svc hitting its max_instances: 64 ceiling in us-east1, which produced repeated infrastructure-initiated cancellations. The rate returned to normal ~15:00 UTC on 8 Sep after the traffic block, with this change still reverted.

Reapplied in !254161 (merged) (19.4). Remaining FE follow-up, independent of this incident: scope adoptRunningStream() to a workflow id so opening a blank chat does not stop an in-flight turn — issue to follow.

Checklist

• Create an issue to reinstate the merge request and assign it to the author of the reverted merge request. → Reinstated directly via !254161 (merged) (author @rfrederico, approved by @ealcantara); no separate issue needed.
• If the revert is to resolve a broken 'master' incident … (n/a)
• If the revert involves a database migration … (n/a)
• Add the appropriate labels before the MR is created.

Related issues and merge requests

• Incident: gitlab-com/gl-infra/production#22875 (closed)
• Reverted: !251389 (merged)
• Reapply: !254161 (merged)
• Original issue: #619497 (closed)