Initial commit: E4B-MarkBase model integration with passing tests

- E4B-MarkBase model (42 layers, 4.4GB) loaded successfully
- All Phase 1-6 tests passed (model loading, forward pass, vision/audio towers, token generation, performance)
- All stress tests passed (5/5 in 127.6s)
  - Concurrent inference
  - Memory stress (67.5 tok/s, 0 NaN)
  - Continuous generation
  - Batch processing
  - Long-running stability
- Swift Metal inference engine with multimodal support

Tests/MarkBaseTests/CoreTests.swift

@@ -0,0 +1,40 @@
+import XCTest
+@testable import MarkBase
+
+final class CoreTests: XCTestCase {
+    
+    func testTokenizerSpaces() throws {
+        let modelDir = "/Users/accusys/MarkBaseEngine/models/E4B-MarkBase"
+        let tokenizer = try TokenizerFactory.load(modelDir: modelDir)
+        
+        // Test space preservation
+        let input = "Hello World"
+        let tokens = tokenizer.encode(text: input)
+        let decoded = tokenizer.decode(tokens: tokens)
+        
+        XCTAssertEqual(decoded.contains(" "), true, "Spaces should be preserved")
+    }
+    
+    func testMultimodalPipeline() throws {
+        let modelDir = "/Users/accusys/MarkBaseEngine/models/E4B-MarkBase"
+        let engine = try MarkBaseEngine(autoCompile: true)
+        let mm = try MultimodalModel(modelDir: modelDir, engine: engine, maxContextLength: 512)
+        
+        XCTAssertNotNil(mm.visionTowerFull, "Vision tower should load")
+        XCTAssertEqual(mm.visionTowerFull?.config.numHiddenLayers, 16, "Vision tower has 16 layers")
+    }
+    
+    func testSamplerFiltering() throws {
+        let sampler = Sampler()
+        
+        // Create test logits
+        var logits = [Float](repeating: 0, count: 262144)
+        logits[258123] = 30.0  // Unused token with high logit
+        logits[500] = 29.0    // Regular token
+        
+        // Test filtering
+        let sampled = sampler.sample(logits: logits, temperature: 1.0, topK: 50, topP: 0.95, filterUnusedTokens: true)
+        
+        XCTAssertLessThan(sampled, 258000, "Should not sample unused tokens when filtering enabled")
+    }
+}