Fix SIGILL in CI: pin Mojo target to x86-64-v3 (AVX2)

The codspeed benchmark runner crashed with exit code 132 (SIGILL)
because `mojo build --emit object` defaults to the native CPU, which
may emit AVX-512 or other instructions the CI runner doesn't support.

Adds MOJO_MCPU env var (defaults to "native") that build.rs passes
as `--mcpu` to the Mojo compiler. CI sets it to "x86-64-v3" (AVX2
baseline) to match the existing RUSTFLAGS="-C target-feature=+avx2".

Signed-off-by: Claude <noreply@anthropic.com>

https://claude.ai/code/session_01EVcJZP4ZmfvWRRg2CsgvST

Author: claude

.github/workflows/ci.yml

@@ -751,6 +751,7 @@ jobs:
       - name: Build benchmarks
         env:
           RUSTFLAGS: "-C target-feature=+avx2"
+          MOJO_MCPU: "x86-64-v3"
         run: cargo codspeed build ${{ matrix.features }} $(printf -- '-p %s ' ${{ matrix.packages }}) --profile bench
       - name: Run benchmarks
         uses: CodSpeedHQ/action@d872884a306dd4853acf0f584f4b706cf0cc72a2

vortex-array/build.rs

@@ -34,10 +34,18 @@ fn main() {
     let obj_path = out_dir.join("vortex_mojo_take.o");
 
     // AOT compile the Mojo kernel to a native object file.
+    //
+    // Use MOJO_MCPU to override the target CPU (defaults to "native"). In CI the runner
+    // CPU may differ from the build host, so we allow pinning to a baseline like
+    // "x86-64-v3" (AVX2) to avoid emitting unsupported instructions (e.g. AVX-512).
+    let mcpu = env::var("MOJO_MCPU").unwrap_or_else(|_| "native".to_owned());
+
     let status = Command::new(&mojo_bin)
         .arg("build")
         .arg("--emit")
         .arg("object")
+        .arg("--mcpu")
+        .arg(&mcpu)
         .arg("-o")
         .arg(&obj_path)
         .arg(&kernel_src)

vortex-array/src/arrays/filter/execute/slice.rs

@@ -6,7 +6,6 @@
 //! Provides both immutable and mutable (in-place) filtering of typed slices by various mask
 //! representations: indices and ranges (slices).
 
-use std::mem::size_of;
 use std::ptr;
 
 use vortex_buffer::Buffer;
@@ -50,11 +49,11 @@ fn filter_slice_by_indices<T: Copy>(slice: &[T], indices: &[usize]) -> Buffer<T>
 
 #[cfg(vortex_mojo)]
 mod mojo {
+    use std::mem::size_of;
+
     use vortex_buffer::Buffer;
     use vortex_buffer::BufferMut;
 
-    use super::size_of;
-
     unsafe extern "C" {
         fn vortex_filter_1byte(src: usize, idx: usize, dst: usize, n: usize);
         fn vortex_filter_2byte(src: usize, idx: usize, dst: usize, n: usize);