fix: vec allocation for AudioRing, catch_unwind on tracing init, profiling

- AudioRing: use vec![].into_boxed_slice() instead of Box::new([]) to
  avoid 32KB stack allocation that crashes scudo on Android
- JNI bridge: wrap tracing_subscriber init in catch_unwind to survive
  sharded_slab allocation failures on some devices
- Engine: per-step encode profiling (avg_agc_us, avg_opus_us, avg_fec_us,
  avg_send_us) logged every 5 seconds in send stats

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

crates/wzp-android/src/audio_ring.rs

@@ -19,7 +19,7 @@ const RING_MASK: usize = RING_CAPACITY - 1;
 
 /// Lock-free single-producer single-consumer ring buffer for i16 PCM samples.
 pub struct AudioRing {
-    buf: Box<[i16; RING_CAPACITY]>,
+    buf: Box<[i16]>,
     /// Monotonically increasing write cursor. ONLY written by producer.
     write_pos: AtomicUsize,
     /// Monotonically increasing read cursor. ONLY written by consumer.
@@ -41,7 +41,7 @@ impl AudioRing {
     pub fn new() -> Self {
         debug_assert!(RING_CAPACITY.is_power_of_two());
         Self {
-            buf: Box::new([0i16; RING_CAPACITY]),
+            buf: vec![0i16; RING_CAPACITY].into_boxed_slice(),
             write_pos: AtomicUsize::new(0),
             read_pos: AtomicUsize::new(0),
             overflow_count: AtomicU64::new(0),