chore: purge 4 dead test files, trim 3 more of dead/slow tests

Delete 4 test files (507 lines):
- release_gate_integration.rs: tests old release gate system that
  was replaced (caused known CI failures)
- apple_silicon_detection_test.rs: all 4 tests gated behind
  cfg(mlx) — MLX is empty [] stub, zero tests ever compiled
- gpu_backend_tests.rs + gpu_layer_verification.rs: all 11 tests
  identical ShimmyUniversalEngine::new() no-ops, no assertions

Trim 3 test files (133 lines removed):
- regression_tests.rs: removed dead no-op tests for MLX (#114)
  and GPU metrics (#111)
- anthropic_api_integration_test.rs: removed test that ran
  cargo build as subprocess (slow, redundant with CI)
- integration_tests.rs: removed 3 #[ignore] no-op tests

All 665 tests pass locally, format clean.

Author: Michael-A-Kuykendall

tests/anthropic_api_integration_test.rs

@@ -3,34 +3,6 @@
 /// This test suite ensures shimmy can serve as a drop-in replacement for
 /// Anthropic's Claude API, enabling tools like Claude Code to work in local networks.
 use serde_json::json;
-use std::process::Command;
-
-#[test]
-fn test_anthropic_api_endpoint_exists() {
-    // Regression test for Issue #109 - Anthropic API format support
-
-    // Build shimmy to ensure anthropic_compat module compiles
-    let output = Command::new("cargo")
-        .args([
-            "build",
-            "--no-default-features",
-            "--features",
-            "huggingface",
-        ])
-        .output()
-        .expect("Failed to build shimmy with anthropic support");
-
-    assert!(
-        output.status.success(),
-        "Failed to build shimmy with Anthropic API support: {}",
-        String::from_utf8_lossy(&output.stderr)
-    );
-
-    // Check that the anthropic_compat module exists in the binary
-    // We can't easily test the HTTP endpoint without starting the server,
-    // but we can verify the module compiles and links correctly
-    println!("✅ Anthropic API compatibility module compiles successfully");
-}
 
 #[test]
 fn test_anthropic_message_format_parsing() {

tests/apple_silicon_detection_test.rs

@@ -36,14 +36,6 @@ async fn create_test_server() -> (String, tokio::task::JoinHandle<()>) {
     (base_url, handle)
 }
 
-#[tokio::test]
-#[ignore] // Requires actual models and Python environment
-async fn test_huggingface_engine_integration() {
-    // This test is disabled until HuggingFace engine is fully implemented
-    // For now we test the interface exists
-    println!("HuggingFace engine test skipped - feature under development");
-}
-
 #[tokio::test]
 async fn test_http_api_health_check() {
     let (base_url, server_handle) = create_test_server().await;
@@ -68,22 +60,6 @@ async fn test_http_api_health_check() {
     server_handle.abort();
 }
 
-#[tokio::test]
-#[ignore] // Requires models to be available
-async fn test_api_generate_endpoint() {
-    // This test is ignored because it would require actual models
-    // Instead, we test the API structure in unit tests
-    println!("API generate test skipped - requires actual models");
-}
-
-#[tokio::test]
-#[ignore] // Requires WebSocket support
-async fn test_websocket_api() {
-    // This test is ignored because it would require full server setup
-    // WebSocket functionality is tested in unit tests
-    println!("WebSocket test skipped - requires full server setup");
-}
-
 #[test]
 fn test_cli_parsing() {
     use clap::Parser;

tests/gpu_backend_tests.rs

@@ -251,29 +251,6 @@ mod regression_tests {
     // Tests for user-reported issues to prevent regressions
     // ========================================
 
-    #[test]
-    fn test_issue_111_gpu_metrics_endpoint() {
-        // Test the fix for Issue #111 - GPU metrics missing from /metrics endpoint
-        use shimmy::engine::adapter::InferenceEngineAdapter;
-        use std::sync::Arc;
-
-        let registry = Registry::default();
-        let engine = Box::new(InferenceEngineAdapter::new());
-        let _state = Arc::new(shimmy::AppState::new(engine, registry));
-
-        // This should not panic and should include GPU fields
-        // Note: We can't easily test the actual HTTP endpoint without starting a server
-        // But we can test that the GPU detection functions work
-
-        // Test that GPU detection functions return valid boolean values
-        // GPU detection functions executed without panicking (test passes if we reach here)
-
-        // In a real test environment, we would test:
-        // 1. GET /metrics returns JSON with gpu_detected field
-        // 2. gpu_vendor field is null or valid vendor string
-        // 3. Fields are properly typed (boolean, string|null)
-    }
-
     #[test]
     fn test_issue_112_safetensors_engine_selection() {
         // Test the fix for Issue #112 - SafeTensors files should use SafeTensors engine
@@ -361,35 +338,6 @@ mod regression_tests {
         assert!(response_json.as_object().unwrap().contains_key("data"));
     }
 
-    #[test]
-    fn test_issue_114_mlx_distribution_features() {
-        // Test the fix for Issue #114 - MLX support in distribution pipeline
-
-        // Test that MLX feature is properly defined
-        #[cfg(feature = "mlx")]
-        {
-            // MLX feature is enabled - test that MLX-related code compiles
-        }
-
-        #[cfg(not(feature = "mlx"))]
-        {
-            // MLX feature is disabled - that's also valid
-        }
-
-        // Test that Cargo.toml includes MLX feature definition
-        let cargo_toml = include_str!("../Cargo.toml");
-        assert!(
-            cargo_toml.contains("mlx = []"),
-            "MLX feature should be defined in Cargo.toml"
-        );
-
-        // Test that Apple Silicon convenience feature includes MLX
-        assert!(
-            cargo_toml.contains("apple = ["),
-            "Apple convenience feature should exist"
-        );
-    }
-
     #[test]
     fn test_cli_model_dirs_option_compatibility() {
         // Test that CLI option doesn't break parsing