fix(ci): align local builds with release target-cpu, drop native flag

The .cargo/config.toml setting target-cpu=native created a dev/release
divergence: local binaries used whatever instructions the dev's host CPU
supported (AVX-512 on modern hardware), while published binaries are
capped at x86-64-v3. CI tests inherited the native flag too, occasionally
producing SIGILL on GH Actions runners whose CPUs reported features the
container couldn't actually execute.

Switch to a per-cfg setting: x86_64 dev builds use target-cpu=x86-64-v3
to match the released Linux/Windows binaries; aarch64 dev builds inherit
rustc defaults to match the released macOS binary. Local dev experience
now reflects what users actually run.

Author: Inuway

.cargo/config.toml

@@ -1,5 +1,14 @@
-# Build for the local CPU architecture (enables AVX2/BMI2/FMA on Ryzen, etc.)
-# Note: this produces a non-portable binary optimized for the dev's CPU.
-# For distribution, override with --target or build without these flags.
-[build]
-rustflags = ["-C", "target-cpu=native"]
+# Local builds use the same target-cpu the CI release workflow uses, so the
+# dev binary has the same instruction set as the binaries users download.
+#
+#   x86_64  (Linux / Windows / Intel Mac): target-cpu=x86-64-v3
+#       AVX2 / BMI1 / BMI2 / FMA baseline, supported by every Intel/AMD CPU
+#       from 2013 onward. Matches the published Linux/Windows release.
+#   aarch64 (Apple Silicon): no override, inherit rustc defaults — matches
+#       the published macOS release, which also sets no target-cpu.
+#
+# Pre-2013 x86 hardware: override this `target-cpu` value (e.g. to
+# `x86-64-v2` or remove the flag entirely) and build from source.
+
+[target.'cfg(target_arch = "x86_64")']
+rustflags = ["-C", "target-cpu=x86-64-v3"]

CONTRIBUTING.md

@@ -12,7 +12,7 @@ cargo build --release
 ./target/release/focalors gui
 ```
 
-The repo's `.cargo/config.toml` sets `target-cpu=native` for local builds — handy when iterating but not portable. The GitHub Actions release workflow builds Linux/Windows binaries with `target-cpu=x86-64-v3` (AVX2 baseline, supported by every Intel/AMD x86 CPU from 2013 onward). macOS builds default to Apple Silicon. Pre-2013 x86 hardware can compile from source instead.
+The repo's `.cargo/config.toml` matches the GitHub Actions release workflow: x86_64 builds (Linux/Windows/Intel Mac) use `target-cpu=x86-64-v3` (AVX2 baseline, supported by every Intel/AMD CPU from 2013 onward); aarch64 builds (Apple Silicon) use rustc defaults. Your local binary has the same instruction set as the binaries users download — what you test is what they run. Pre-2013 x86 hardware can compile from source by lowering `target-cpu` in `.cargo/config.toml`.
 
 ## Engine development