fix: Simplify to CUDA-only GPU support for testing

Author: Michael-A-Kuykendall

.github/workflows/release.yml

@@ -57,13 +57,6 @@ jobs:
         uses: Jimver/cuda-toolkit@v0.2.14
         with:
           cuda: '12.4'
-          
-      - name: Install Vulkan SDK  
-        if: contains(matrix.features, 'gpu') || contains(matrix.features, 'llama-vulkan')
-        uses: humbletim/install-vulkan-sdk@v1.2
-        with:
-          version: 1.3.231.1
-          cache: true
 
       - name: Install system dependencies (Linux)
         if: runner.os == 'Linux'

CLAUDE.md

@@ -7,7 +7,7 @@ Shimmy is a lightweight sub-20MB Rust inference engine serving as an optimal shi
 
 **READ BEFORE WRITE**: Always read a file before editing or writing to it (Claude Code requirement)
 **FOLLOW INTEGRATION PLAN**: Check integration plans before implementation decisions
-**PUBLIC RELEASE APPROVAL**: Human approval required for releases, Cargo.toml, README changes
+**MANDATORY RELEASE APPROVAL**: NEVER create releases, tags, or version bumps without explicit human authorization. Always ask "Should I proceed with creating release v[X.X.X]?" before any release actions
 **PROFESSIONAL LANGUAGE**: No profanity, maintain professional standards
 **CONVENTIONAL COMMITS**: Use conventional commit format for all commits