refactor: Extract `turborepo-query-api` trait crate for compile-time decoupling (#12165)
## Summary
- Introduces `turborepo-query-api`, a trait crate defining `QueryServer`
and `QueryRun` that decouples `turborepo-lib` from the heavy
`turborepo-query` crate (async-graphql, axum, oxc), enabling parallel
compilation
- The binary crate wires the two halves together via `TurboQueryServer`,
implementing the dependency inversion pattern
## Key design decisions
**Error composition over duplication.** Rather than duplicating error
variants between crates with a fragile manual `From` impl,
`turborepo-query::Error` wraps `turborepo_query_api::Error` via an
`Api(#[from] ...)` variant. Shared errors (Boundaries, Server, Path, UI,
etc.) flow through automatically. Query-specific errors (Trace,
FileNotFound, Serde, Parse) stay local and get boxed only when crossing
the boundary.
**Resolution errors preserve diagnostics.** Added `Resolution` variant
with `#[diagnostic(transparent)]` to the API error enum so miette source
spans survive the full error chain from query → API → CLI.
**Watch mode threading.** `WatchClient` now accepts and forwards
`query_server` to all 3 `RunBuilder` call sites, fixing a silent Web UI
regression where `turbo watch --ui=web` would produce no UI and no
error.
**Unified error type on `QueryServer`.** All three trait methods now
return `Result<_, turborepo_query_api::Error>` instead of mixing in
`turborepo_ui::Error`.
## How to test
1. `cargo check -p turbo --features rustls-tls` — verifies compilation
2. `cargo clippy -p turborepo-query-api -p turborepo-query -p
turborepo-lib -p turbo --features rustls-tls` — zero warnings
3. Run `turbo query "{ packages { items { name } } }"` in any monorepo —
should work as before
4. The `QueryNotAvailable` error can be verified by inspecting the code
path (only reachable if someone uses `turborepo-lib` as a library
without providing a query server)
turbo
7acfdc43 - refactor: Extract `turborepo-query-api` trait crate for compile-time decoupling (#12165)
