Replace hardcoded DSML typo variant lists in Go/Node tool call parsers with
generalized prefix consumption that tolerates repeated leading <, repeated DSML
prefix noise, and trailing pipe terminators. Split tiktoken-dependent token
counting into a build-tagged file for non-cgo platform compatibility. Add /data
directory to Dockerfile for bind-mount permissions.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This change reorganizes the codebase for better maintainability and performance while preserving API surface.
- Create ds2api package with modules: core, services, utils, models
- Migrate config, logging, auth, DeepSeek, PoW, and message processing into dedicated modules
- Introduce PoW caching (60s TTL) and async/parallel support (scalability for multiple requests)
- Replace direct curl calls with a pool-enabled HTTP client setup and WASM-based PoW engine
- Add in-memory token/account management scaffolding and improved token estimation
- Optimize streaming paths and prepare for better backpressure and concurrency
- Remove transformers/tokenizer usage and keep a simple token length estimator
Non-breaking migration: keep API endpoints intact; new structure under ds2api is transparent to clients
