debug: set ASAN_OPTIONS=protect_shadow_gap=0 for GPU compress fuzz

The compress_gpu fuzz job aborts on the very first input with a "deadly
signal": CudaSession::try_default() returns Err and .expect("Failed to
initialize CUDA device 0") panics (vortex-cuda/src/session.rs:168). It is
input-independent (even the empty seed trips it) because the session is a
LazyLock initialized on first use.

cargo-fuzz builds with AddressSanitizer, which mprotects the shadow gap
PROT_NONE. That collides with the large virtual-address range the CUDA
driver reserves, so device init fails with an OOM-style error. Setting
protect_shadow_gap=0 stops ASan protecting the gap and frees the VA range
for CUDA.

Temporary debugging change on this branch; revert with the rest once the
GPU fuzz job is healthy.

Signed-off-by: Andrew Duffy <andrew@a10y.dev>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: a10y

.github/workflows/fuzz.yml

@@ -207,6 +207,10 @@ jobs:
       fuzz_target: compress_gpu
       runner: gpu
       extra_features: "cuda"
+      # ASan mprotects the shadow gap PROT_NONE by default, which collides with the large
+      # virtual-address range the CUDA driver reserves and makes device init fail with an
+      # OOM-style error. protect_shadow_gap=0 frees that VA range for CUDA.
+      extra_env: "ASAN_OPTIONS=protect_shadow_gap=0"
       jobs: 1
     secrets:
       R2_FUZZ_ACCESS_KEY_ID: ${{ secrets.R2_FUZZ_ACCESS_KEY_ID }}