fix(server): retain OpenCode V2 per-turn token usage

[saphid]

OpenCode V2 now records per-turn token usage from step-finish events, deduplicating step IDs and carrying input, cache, output and reasoning totals into the turn. Completed turns report complete usage, interrupted/failed turns preserve partial usage, and turns without usage data report unavailable rather than zero. Subagent presence is tracked separately from the main-agent totals.

Current head ea21a4983 includes current V2 base 3aa47e294 and the shared test-only prerequisite #10865. Merge #10865 first; the topic itself changes 2 files. This rebuild removes the obsolete merged-base history from the PR diff.

Validation: 58 focused tests pass across the affected feature and foundation persistence coverage. Scoped server typecheck passes; targeted lint, formatting and diff checks pass. After merging the latest V2 base, 24 OpenCode adapter and targeted queued-cancellation replay tests pass; server typecheck passes. New-head CI is pending.

Independent cross-provider review was unavailable: the direct Claude Opus 5 high launcher in this audit exited 1 with expired OAuth before inference. No Claude review is claimed.

Targets Orchestrator V2 in #2829. Updated with GPT-6 in Codex/T3.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This focused fix restores per-turn token accounting and changes provider-turn usage data consumed by analytics and potentially metering. The implementation is isolated and tested, but the billing/usage surface warrants human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[cursor]

Bugbot is paused — on-demand spend limit reached

Bugbot uses usage-based billing for this team and has hit its on-demand spend limit.

A team admin can raise the spend limit in the Cursor dashboard, or wait for the next billing cycle to continue.

[saphid]

Independent automated first-pass review (Amp/Astra agent mode) against base 415ed0f: no actionable findings. Verified in source: usage normalization matches the pinned OpenCode v1.15.13 conversion; step-ID dedup handles replay; root/child accumulation stays separate; unavailable/partial shapes satisfy the TurnTokenUsage contract; hasSubagents is only set on task tools; tokens fields are SDK-required and defensively clamped. One note: the PR-side diff vs the current base tip is inflated because the base branch advanced past 415ed0f; the contribution itself remains the two adapter files.

[cursor]

Bugbot is paused — on-demand spend limit reached

Bugbot uses usage-based billing for this team and has hit its on-demand spend limit.

A team admin can raise the spend limit in the Cursor dashboard, or wait for the next billing cycle to continue.

[macroscopeapp]

Macroscope has since reviewed this pull request. An earlier review was skipped by a cost limit; a review has now completed, so that notice no longer applies.

[juliusmarminge]

Closing as superseded by 95045bb, already on remote v2. The adapter retains normalized per-turn OpenCode step-finish usage, deduplicates step IDs, reports main-agent scope and subagent presence, and distinguishes complete/partial/unavailable totals. Current code additionally checks message ownership to avoid charging unrelated/replayed steps to a turn. All five focused usage cases pass on current v2.