feat(cw): add CwDeepDecoder wiring ONNX Runtime to the CW pipeline

组装完整解码链路: 麦克风 PCM -> 重采样 3200Hz -> 20 秒滚动缓冲 ->
频谱图 -> ONNX 推理 -> 贪心 CTC -> 文本。

ICwDecoder 放 core:domain (纯 Kotlin, UI 可直接依赖), 实现放 core:data
(需要 Android Context 读 assets 及 ONNX Runtime 原生库)。

关键设计:
- decodedText 是替换语义不是追加: 模型会随上下文增加改写先前字符, 追加
  会把中间态 (BM -> BG7 -> BG7NTA) 永久留在屏幕上
- 推理用 Mutex.tryLock 串行化: 上一次未完成时直接跳过本周期, 慢设备不会
  堆积任务导致 OOM
- 推理在 Dispatchers.Default, 不阻塞音频采集线程
- 模型元数据从 model.onnx.json 读取 (字符表/blank 索引/输入输出名), 不在
  代码里硬编码, 模型更新时无需改代码
- lastInferenceMs 打点: 手机实际推理耗时需真机实测 (估算系数待验证)
- estimatedPitch 由频谱峰值 bin 反算, 仅供 UI 显示 —— 模型 400-1200Hz
  固定窗自带定频, 不需要频谱峰值跟踪参与解码
- 模型加载失败时 errorMessage 置位 (旧内核已删, 无兜底可退)
- close() 释放 OrtSession, 否则原生内存泄漏

验证:
./gradlew :core:data:compileDebugKotlin => BUILD SUCCESSFUL

core/domain/src/main/java/com/rtbishop/look4sat/core/domain/cw/ICwDecoder.kt

@@ -0,0 +1,59 @@
+/*
+ * Look4Sat. Amateur radio satellite tracker and pass predictor.
+ * Copyright (C) 2019-2026 Arty Bishop and contributors.
+ *
+ * This program is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+package com.rtbishop.look4sat.core.domain.cw
+
+import kotlinx.coroutines.flow.StateFlow
+
+/**
+ * A CW (Morse code) decoder fed with microphone PCM.
+ *
+ * Implementations live outside `core:domain` when they need platform APIs;
+ * this contract stays pure Kotlin so the UI can depend on it directly.
+ */
+interface ICwDecoder {
+
+    /**
+     * Decoded text for display.
+     *
+     * Note this is **replace** semantics, not append: a whole-segment model
+     * revises earlier characters as more audio arrives, so consumers must show
+     * the current value rather than accumulating emissions.
+     */
+    val decodedText: StateFlow<String>
+
+    /** Detected tone frequency in Hz, or null before a tone is found. */
+    val estimatedPitch: StateFlow<Float?>
+
+    /** Relative signal strength in 0..1 for level meters. */
+    val signalStrength: StateFlow<Float>
+
+    /** Most recent inference duration in milliseconds, for diagnostics. */
+    val lastInferenceMs: StateFlow<Int>
+
+    /** Non-null when the decoder cannot run, for example the model failed to load. */
+    val errorMessage: StateFlow<String?>
+
+    /** Feed captured mono PCM in -1..1. Safe to call from a capture thread. */
+    suspend fun processBuffer(samples: FloatArray, sampleRate: Int)
+
+    /** Clear decoded text and buffered audio. */
+    fun reset()
+
+    /** Release native resources. Must be called when the decoder goes away. */
+    fun close()
+}