gen_latch: Synchronize at start to ensure broadcast of initial "not written" state (nw)

This fixes sound in onna34ro and makes the nycaptor hack unnecessary. Occasional sound loss in flstory remains, but that driver is IMPERFECT_SOUND anyway; I suspect the fault there may lie with the emulation of the MSM5232 or its interface to the Z80 here.

src/devices/machine/gen_latch.cpp

@@ -42,6 +42,18 @@ void generic_latch_base_device::device_start()
 {
 	m_data_pending_cb.resolve_safe();
 	save_item(NAME(m_latch_written));
+
+	// synchronization is needed since other devices may not be initialized yet
+	machine().scheduler().synchronize(timer_expired_delegate(FUNC(generic_latch_base_device::init_callback), this));
+}
+
+//-------------------------------------------------
+//  init_callback - set initial state
+//-------------------------------------------------
+
+void generic_latch_base_device::init_callback(void *ptr, s32 param)
+{
+	m_data_pending_cb(m_latch_written ? 1 : 0);
 }
 
 //-------------------------------------------------

src/devices/machine/gen_latch.h

@@ -67,6 +67,8 @@ protected:
 	void set_latch_written(bool latch_written);
 
 private:
+	void init_callback(void *ptr, s32 param);
+
 	bool                    m_separate_acknowledge;
 	bool                    m_latch_written;
 	devcb_write_line        m_data_pending_cb;

src/mame/drivers/nycaptor.cpp

@@ -733,9 +733,6 @@ void nycaptor_state::machine_start()
 	save_item(NAME(m_char_bank));
 	save_item(NAME(m_palette_bank));
 	save_item(NAME(m_gfxctrl));
-
-	// hack to put input_merger in correct initial state
-	m_soundnmi->in_w<0>(0);
 }
 
 void nycaptor_state::machine_reset()