Collapse discovery TXT into one seam in usbip-core — Rust + Kotlin reuse

[stanvx]

Parent

Architecture review (2026-06-21). Slice 4 of 6 from the deepening series. Documented at server/usbip-server/src/discovery.rs, client/usbip-client/src/discovery.rs, and server/usbip-server/src/api.rs:100-103.

What to build

The wire format for mDNS TXT records is encoded in server/usbip-server/src/discovery.rs (encode_devices_txt + 3 tests) and decoded in client/usbip-client/src/discovery.rs (decode_devices_txt + 0 tests). Two implementations of one wire format with no shared contract. The future Kotlin client will be the third.

Lift both into a new usbip_core::discovery_txt module. Server encodes via that module. Client decodes via that module. Treat the wire format as a contract with golden fixtures. Add a #[doc(hidden)] const TXT_FORMAT_V1: &[u8] = ... so the format version is visible at the call site.

This is the smallest, highest-leverage contract change. Once shared, the Kotlin client cannot drift from the Rust encoder.

Acceptance criteria

• New module usbip-core/src/discovery_txt.rs with pub fn encode(devices: &[DeviceDescriptor]) -> String and pub fn decode(txt: &str) -> Result<Vec<DeviceDescriptor>, DiscoveryTxtError>
• Server's encode_devices_txt becomes a thin re-export or is deleted
• Client's decode_devices_txt becomes a thin re-export or is deleted
• At least 6 round-trip tests cover: empty device list, single device, multiple devices, special characters in product strings, malformed input rejection, format version mismatch
• tests/fixtures/discovery_txt/ contains a golden TXT file and the test asserts byte-for-byte encoding stability
• Module doc-comment records the wire format with a BNF-style grammar
• cargo test --workspace --release passes; cargo clippy --workspace -- -D warnings clean

Blocked by

• Reconcile docs to code — name module ownership in CONTEXT.md, fix 12+ divergences #27 (Reconcile docs to code) — needs DiscoveryTxtOwner line in CONTEXT.md before lift