Phase 3: dictionary loading pipeline + binary embedding

- engine_init.cc: loads StenoDictionaryCollection from embedded binary,
  validates JSC4 magic + timestamp, constructs dict list + full processor
  pipeline (Repeat → AllUp → JeffModifiers → Engine)
- dict_embed.S: embeds compiled dict binary via .incbin directive.
  Falls back to null placeholder if no dict available
- CMakeLists.txt: dict binary discovery (dicts/steno_dict.bin or
  JAVELIN_DICT_BIN cmake var), assembly embedding
- tools/dict_compiler/compile_dict.py: stroke parser + skeleton compiler.
  Full JSC4 binary format not yet implemented — for now, use Javelin
  web tool (lim.au) to generate dict binary
- dicts/: directory for compiled dictionary binaries

Dict compiler is the remaining gap — JSC4 format requires exact XIP
pointer layout matching Javelin's C++ struct casting.

src/dict_embed.S

@@ -0,0 +1,17 @@
+/* Embed compiled steno dictionary binary into firmware.
+ * JAVELIN_DICT_BIN_PATH is set by CMake to the path of the compiled
+ * dictionary file. If not set, a 4-byte null placeholder is used.
+ */
+
+.section .rodata.steno_dict,"a",%progbits
+.global _javelin_dict_start
+.global _javelin_dict_end
+.balign 4
+
+_javelin_dict_start:
+#ifdef JAVELIN_DICT_BIN_PATH
+.incbin JAVELIN_DICT_BIN_PATH
+#else
+.byte 0x00, 0x00, 0x00, 0x00
+#endif
+_javelin_dict_end:

src/engine_init.cc

@@ -1,8 +1,9 @@
 #include "zmk_javelin_steno/zmk_platform_shim.h"
 
+#include "container/list.h"
 #include "dictionary/dictionary_definition.h"
 #include "dictionary/dictionary_list.h"
-#include "dictionary/invalid_dictionary.h"
+#include "dictionary/user_dictionary.h"
 #include "engine.h"
 #include "orthography.h"
 #include "processor/all_up.h"
@@ -14,45 +15,86 @@
 #include "static_allocate.h"
 #include "stroke.h"
 
-// Pipeline: StenoProcessor → StenoRepeat → StenoAllUp → StenoJeffModifiers → StenoEngine
-//
-// StenoProcessor converts raw StenoKey press/release → StenoKeyState.
-// StenoRepeat handles stroke repetition on held keys.
-// StenoAllUp triggers when all keys released (standard steno behavior).
-// StenoJeffModifiers handles modifier key combos within steno.
-// StenoEngine does dictionary lookup → text output via Key::Press/Release.
+extern "C" {
+#include <zephyr/kernel.h>
+#include <zephyr/logging/log.h>
+}
+
+LOG_MODULE_DECLARE(zmk, CONFIG_ZMK_LOG_LEVEL);
+
+// Linker symbols for the embedded dictionary binary.
+// Defined by the linker script or incbin directive in dict_embed.S
+extern "C" const uint8_t _javelin_dict_start[];
+extern "C" const uint8_t _javelin_dict_end[];
 
 static JavelinStaticAllocate<StenoCompiledOrthography> compiledOrthography;
+static JavelinStaticAllocate<StenoDictionaryList> dictionaryList;
 static JavelinStaticAllocate<StenoJeffModifiers> jeffModifiers;
-static JavelinStaticAllocate<StenoRepeat> repeat;
+static JavelinStaticAllocate<StenoRepeat> stenoRepeat;
 static JavelinStaticAllocate<StenoAllUp> allUp;
 static StenoProcessor *processor = nullptr;
 
 static bool initialized = false;
 
+static bool load_dictionary_collection() {
+  const auto *collection =
+      reinterpret_cast<const StenoDictionaryCollection *>(_javelin_dict_start);
+
+  if (collection->magic != STENO_MAP_DICTIONARY_COLLECTION_MAGIC) {
+    LOG_ERR("Steno dict magic mismatch: 0x%08x (expected 0x%08x)",
+            collection->magic, STENO_MAP_DICTIONARY_COLLECTION_MAGIC);
+    return false;
+  }
+
+  if (!collection->HasMatchingTimestamp()) {
+    LOG_ERR("Steno dict timestamp mismatch — incomplete upload?");
+    return false;
+  }
+
+  List<StenoDictionaryListEntry> entries;
+  collection->AddDictionariesToList(entries);
+
+  if (entries.IsEmpty()) {
+    LOG_ERR("Steno dict collection has no dictionaries");
+    return false;
+  }
+
+  new (dictionaryList) StenoDictionaryList(static_cast<List<StenoDictionaryListEntry>&&>(entries));
+
+  const StenoOrthography *ortho = &StenoOrthography::emptyOrthography;
+  // TODO: load orthography from collection if present
+  new (compiledOrthography) StenoCompiledOrthography(*ortho);
+
+  new (StenoEngine::container)
+      StenoEngine(dictionaryList.value, nullptr, compiledOrthography.value);
+
+  LOG_INF("Steno engine initialized with %d dictionaries",
+          collection->dictionaryCount);
+  return true;
+}
+
 extern "C" {
 
 void zmk_javelin_steno_init(void) {
   if (initialized) {
     return;
   }
-
-  // Use empty orthography until a dictionary collection is loaded (Phase 3).
-  // The empty orthography has no rules — suffix folding won't work, but
-  // basic dictionary lookups will function once a dict is in flash.
-  new (compiledOrthography)
-      StenoCompiledOrthography(StenoOrthography::emptyOrthography);
-
-  // TODO Phase 3: Load StenoDictionaryCollection from flash partition.
-  // 1. Get pointer to steno_dict_partition start address
-  // 2. Validate magic == 0x3443534a ('JSC4')
-  // 3. Call collection->AddDictionariesToList() to populate dict list
-  // 4. Construct engine with real dictionary
-  //
-  // For now, engine construction is deferred until dictionary is available.
-  // The processor pipeline is NOT built yet — process_key will early-return.
-
   initialized = true;
+
+  if (!load_dictionary_collection()) {
+    LOG_WRN("Steno engine not started — no valid dictionary found");
+    return;
+  }
+
+  // Build processor pipeline: Repeat → AllUp → JeffModifiers → Engine
+  new (jeffModifiers) StenoJeffModifiers(StenoEngine::container.value);
+  new (allUp) StenoAllUp(jeffModifiers.value);
+  new (stenoRepeat) StenoRepeat(allUp.value);
+
+  static StenoProcessor processorInstance(stenoRepeat.value);
+  processor = &processorInstance;
+
+  LOG_INF("Steno processor pipeline ready");
 }
 
 void zmk_javelin_steno_process_key(int steno_key_index, bool is_press) {
@@ -63,19 +105,3 @@ void zmk_javelin_steno_process_key(int steno_key_index, bool is_press) {
 }
 
 } // extern "C"
-
-// Called once dictionary is loaded (Phase 3) to complete initialization.
-void zmk_javelin_steno_init_engine(StenoDictionary &dictionary,
-                                   const StenoOrthography &orthography) {
-  new (compiledOrthography) StenoCompiledOrthography(orthography);
-
-  new (StenoEngine::container)
-      StenoEngine(dictionary, nullptr, compiledOrthography.value);
-
-  new (jeffModifiers) StenoJeffModifiers(StenoEngine::container.value);
-  new (allUp) StenoAllUp(jeffModifiers.value);
-  new (repeat) StenoRepeat(allUp.value);
-
-  static StenoProcessor processorInstance(repeat.value);
-  processor = &processorInstance;
-}