perf(asr): speed up local Flow dictation and land CLM/keyboard refactor

Reduce perceived latency from key release to final text:
- Adaptive chunking: 2.5s first chunk + 5s follow-ups so short
  utterances start on-device recognition while still recording.
- Session-level ASR warmup and audio-format cache reuse to remove
  per-utterance cold-start of SpeechAnalyzer.
- Mirror live pipelined partials to the keyboard transcript line via
  a new flow.transcriptionPartial App Group key + Darwin ping.

Also commits the accumulated custom language model, Flow session,
keyboard extension restructure, and Xiaomi MiMo provider work in
progress on this branch.

OSGKeyboardShared/Utilities/OSGLog.swift

@@ -0,0 +1,17 @@
+// OSGLog.swift
+// OSGKeyboard · Shared
+//
+// Unified os.Logger categories for cross-target diagnostics. Filter in
+// Console.app with subsystem `com.osgkeyboard.ios`.
+
+import os
+
+public enum OSGLog {
+    private static let subsystem = "com.osgkeyboard.ios"
+
+    public static let flow = Logger(subsystem: subsystem, category: "flow")
+    public static let clm = Logger(subsystem: subsystem, category: "clm")
+    public static let config = Logger(subsystem: subsystem, category: "config")
+    public static let asr = Logger(subsystem: subsystem, category: "asr")
+    public static let keyboardExt = Logger(subsystem: subsystem, category: "keyboardExt")
+}

OSGKeyboardShared/Utilities/UtteranceStreamChunker.swift

@@ -17,7 +17,8 @@ public enum UtteranceStreamChunker {
         AsyncStream { continuation in
             let task = Task {
                 var buffer: [Float] = []
-                buffer.reserveCapacity(config.maxChunkSamples + config.pauseExtensionSamples)
+                let initialCapacity = config.maxChunkSamples(forChunkIndex: 0) + config.pauseExtensionSamples
+                buffer.reserveCapacity(initialCapacity)
                 var chunkIndex = 0
 
                 func emit(upTo splitEnd: Int, isLast: Bool) {
@@ -40,8 +41,12 @@ public enum UtteranceStreamChunker {
                     guard !snap.samples.isEmpty else { continue }
                     buffer.append(contentsOf: snap.samples)
 
-                    while buffer.count >= config.maxChunkSamples {
-                        let split = pauseAwareSplitIndex(in: buffer, config: config)
+                    while buffer.count >= config.maxChunkSamples(forChunkIndex: chunkIndex) {
+                        let split = pauseAwareSplitIndex(
+                            in: buffer,
+                            config: config,
+                            chunkIndex: chunkIndex
+                        )
                         emit(upTo: split, isLast: false)
                     }
                 }
@@ -66,9 +71,10 @@ public enum UtteranceStreamChunker {
     /// Pick a split index at or after `maxChunkSamples`, preferring a pause.
     static func pauseAwareSplitIndex(
         in buffer: [Float],
-        config: FlowUtteranceChunkConfig
+        config: FlowUtteranceChunkConfig,
+        chunkIndex: Int = 1
     ) -> Int {
-        let minSplit = config.maxChunkSamples
+        let minSplit = config.maxChunkSamples(forChunkIndex: chunkIndex)
         guard buffer.count >= minSplit else { return buffer.count }
 
         let searchEnd = min(buffer.count, minSplit + config.pauseExtensionSamples)