xray-rust 0.8.1

Stable patch release of xray-rust 0.8.1. The mobile SDK is distributed separately through xray-rust-mobile.

Changes:

• Fix Apple VPN startup connectivity by running the startup probe after DNS bootstrap and before installing tunnel routes/DNS. Failure, timeout or cancellation leaves network settings untouched (#70).
• Add C ABI 1.12 and STARTUP_PROBE_PREFLIGHT, with matching Swift adapter support. The updated Apple provider requires the matching native library when startup probes are enabled; protocol/configuration formats are unchanged.
• Update merged Rust dependencies and core Android build tools. Include UDP/WireGuard fixture corrections and CI/interoperability maintenance.

Known limitations explicitly accepted for 0.8.1:

• SS2022 passed physical iPhone Wi-Fi → cellular → Wi-Fi and lock/wake only with IP_MTU_DISCOVER=0 on the owned reference-server IPv4 UDP socket. The default-policy baseline failed on a 1,372-byte logical IPv6 UDP echo. This is conditional deployment evidence, not a client or default-path fix.
• Four Android SS2022 remote EOF observations exceeded the 3-second bound across FD/packet-pump and close/disconnect. Local cancellation and subsequent traffic recovery passed; bounded remote EOF is not claimed.
• The initial SS2022 ChaCha20 p95 latency comparison exceeded 5%. A predeclared 15-pair follow-up did not reproduce it, but its combined uncertainty interval crosses 5%. A 5% non-regression bound remains unconfirmed.
• Android network-switch and lock/wake checks retain their existing v0.8 owner-skipped scope and remain not tested. Existing unsupported combinations and documented protocol/resource tradeoffs remain applicable.

Validation includes fresh exact-candidate physical iPhone/Samsung measurements, complete core/SDK candidate CI, and 13 schema-4 performance metrics. Trojan and VMess passed iPhone WAN/lock checks with default server settings. All successful WAN runs retained one runtime ID; they test fresh-flow recovery, not established TCP-session survival. The Android functional/resource runs passed with all observed traffic delivered. These measurements do not close historical issues or establish universal network/performance parity.

Original failures, negative controls, fixture conditions, exact binary/source identities and the explicit 0.8.1 owner decision are retained in the attached evidence. A schema-validation pass does not mean every trial passed.

Measured report · Exact-candidate evidence validation.

Source verification: immutable tag checks confirm every other required tag CI gate and two independent secret scans. The initial tag CI secret-scan failure is retained: it flagged six verified public binary SHA-256 values on a separate evidence branch. The release tag was not changed.