fix: preserve interrupted reasoning and improve usage settlement

[chenyme]

Summary

Fix interrupted model generations losing streamed reasoning content and reporting incomplete token usage.

This change:

• Flushes and persists buffered reasoning with an independent settlement context after cancellation.
• Prevents canceled request contexts from blocking terminal trace persistence.
• Preserves the richer live reasoning trace on the frontend for interrupted responses only.
• Uses upstream usage when available and estimates missing output/reasoning usage from visible content.
• Records usage provenance as observed, recovered, estimated, or mixed.
• Retains estimated input usage from earlier tool-loop calls when the current Responses API usage is recovered.
• Adds an SQLite-backed integration test covering cancellation, persistence, and trace reload.

Change type

• Bug fix
• Feature
• Documentation
• Refactor
• Configuration / deployment
• Security hardening
• Other

Affected areas

• Frontend / UI
• Backend / API
• Authentication / authorization
• Conversations / streaming
• Files / RAG / extraction
• Model routing / providers
• MCP / tools
• Billing / payments
• Admin console
• Deployment / Docker / configuration
• Documentation

Verification

• pnpm --filter @deeix/web check
• pnpm --filter @deeix/web build
• cd backend && go test ./internal/application/conversation ./internal/application/billing -count=1
• git diff --check HEAD
• Verified cancellation against the locally configured PostgreSQL LongCat-2.1 model.

Screenshots, API examples, or logs

Local LongCat-2.1 cancellation verification:

• Reasoning received before cancellation: 80 bytes
• Persisted reasoning after reload: 80 bytes
• Estimated reasoning usage: 18 tokens
• Usage source: estimated

The upstream did not return final usage after cancellation, so visible-content estimation was used as the fallback.

Configuration, migration, and compatibility notes

• No database migration is required.
• No public API contract or Swagger changes are required.
• Existing successful generation behavior is unchanged.
• Recovered upstream output and reasoning usage remains authoritative.
• Token estimates are used only when final upstream usage is unavailable or incomplete.
• Usage provenance is stored in the existing pricing snapshot JSON.

Documentation

• Documentation is not needed for this change.
• Documentation was updated.
• Documentation still needs to be updated.

Security and privacy

• No secrets, tokens, credentials, local config, or personal data are included.
• User data access remains scoped by authenticated user context unless an admin-only path explicitly requires broader access.
• Security-sensitive behavior was reviewed, including authentication, authorization, provider routing, file processing, billing, and admin APIs where relevant.

Checklist

• I searched existing issues and pull requests.
• Changes are focused and do not include unrelated refactors.
• Tests or static verification were run where practical.
• User-facing behavior, deployment steps, API contracts, or configuration changes are documented.
• Generated artifacts are included only when this project explicitly requires them.
• Caches, build output, .pyc files, .env files, and local storage data are not committed.