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

Sources/MarkBase/Vision/VisionConfig.swift

@@ -0,0 +1,45 @@
+import Foundation
+
+public struct VisionConfig: Codable {
+    public let hiddenSize: Int
+    public let numAttentionHeads: Int
+    public let numHiddenLayers: Int
+    public let headDim: Int
+    public let globalHeadDim: Int
+    public let intermediateSize: Int
+    public let hiddenAct: String
+    public let rmsNormEps: Float
+    public let outputProjDims: Int
+    public let patchSize: Int
+    public let imageSize: Int
+    
+    public init(
+        hiddenSize: Int = 768,
+        numAttentionHeads: Int = 12,
+        numHiddenLayers: Int = 12,
+        headDim: Int = 64,
+        globalHeadDim: Int = 64,
+        intermediateSize: Int = 3072,
+        hiddenAct: String = "gelu_pytorch_tanh",
+        rmsNormEps: Float = 1e-6,
+        outputProjDims: Int = 1536,
+        patchSize: Int = 14,
+        imageSize: Int = 224
+    ) {
+        self.hiddenSize = hiddenSize
+        self.numAttentionHeads = numAttentionHeads
+        self.numHiddenLayers = numHiddenLayers
+        self.headDim = headDim
+        self.globalHeadDim = globalHeadDim
+        self.intermediateSize = intermediateSize
+        self.hiddenAct = hiddenAct
+        self.rmsNormEps = rmsNormEps
+        self.outputProjDims = outputProjDims
+        self.patchSize = patchSize
+        self.imageSize = imageSize
+    }
+    
+    public var numPatches: Int {
+        (imageSize / patchSize) * (imageSize / patchSize)
+    }
+}