CI: disable AVX512-FP16 host codegen on the Aurora LTS self-runner

Under concurrent oneMKL load on the Aurora Sapphire Rapids nodes, Julia's native
AVX512-FP16 host codegen silently miscomputes Float16 (e.g. the GPUArrays
`A .* B .+ c` broadcast reference used by gpuarrays/broadcasting), failing tests
even though the GPU result is correct. It is single-process-clean, MXCSR-clean,
and only the native-FP16 path is affected (the Float32 path is immune).

Run the test suite with `-C native,-avx512fp16`, which routes Float16 through
Float32 and propagates to the Pkg.test subprocess and its parallel Malt workers
via Base.julia_cmd(). `julia-actions/julia-runtest` cannot pass a cpu-target, so
invoke `Pkg.test()` directly. A/B under the reproducer: native FP16 -> ~6% of
checks corrupt; -avx512fp16 -> 0 / 35000.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: michel2323

.github/workflows/ci.yml

@@ -36,5 +36,13 @@ jobs:
       - uses: julia-actions/cache@v3
       - uses: julia-actions/julia-buildpkg@latest
         continue-on-error: true
-      - uses: julia-actions/julia-runtest@latest
+      # Disable AVX512-FP16 host codegen on the Aurora Sapphire Rapids nodes. Under concurrent
+      # oneMKL load the native AVX512-FP16 path silently miscomputes *host* Float16 (e.g. the
+      # GPUArrays `A .* B .+ c` broadcast reference), failing tests even though the GPU result
+      # is correct (single-process clean; MXCSR clean; only the native-FP16 path, not Float32).
+      # `-C native,-avx512fp16` routes Float16 through Float32 and propagates to the Pkg.test
+      # subprocess and its parallel workers via Base.julia_cmd(). `julia-runtest` cannot pass a
+      # cpu-target, so invoke Pkg.test() directly. See repro_bcast_mkl.jl.
+      - name: Run tests (AVX512-FP16 disabled)
         continue-on-error: true
+        run: julia -C "native,-avx512fp16" --color=yes --project=. -e 'import Pkg; Pkg.test()'