Send billing reports to Thalamus for vault-served requests
Per-request billing reports through Thalamus's routing accessor, for vault-served requests on OpenCode 2 (HTTP and xAI WebSocket) and Pi; OpenCode 1 sends none. Without the accessor no billing work runs. Provider usage semantics, charge basis and automatic-cache claims are a cited table in packages/core/src/providers.ts and docs/billing-reports.md; rows marked assumed are checked against live replies before release.
Summary by cubic
Sends per-request billing reports to Thalamus via the routing accessor for vault-served requests on OpenCode 2 HTTP, xAI WebSocket, and Pi; OpenCode 1 sends none. Billing only runs for placeholder-key requests with a vault receipt and an installed accessor. Provider usage semantics and charge basis live in a cited table in packages/core/src/providers.ts and docs/billing-reports.md.
New Features
- Retries and account reroutes report each physical send with its own UUID and attempt number; redirects core follows are routing hops, not billable calls.
- Pi reports provider-wire usage with its final stored responseId; cancellation after a protocol terminal marker reports completed, while pre-terminal cuts report cut.
- A live-smoke mode for OpenCode 2 checks cache and reasoning reports against host-stored counters, using fresh identical-prompt sessions for the cache pair and stored visible output for reasoning.