OpenCode 1.18.17 tightens agent reliability

This is practical infrastructure work for an AI coding agent: fewer silent failures, more predictable sessions, and cleaner operational visibility.

• –Server-default model syncing prevents stale local configuration from unexpectedly steering sessions
• –Bounded retries reduce runaway failures and duplicate requests
• –Better compaction context retention should improve continuity across long coding tasks
• –Go quota visibility gives developers clearer insight into usage limits
• –R2 SQL-backed stats suggest a more scalable analytics path as OpenCode usage grows