Remove per-packet backpressure: streaming 3x faster (#35)

With COBS trailing zero bug fixed + D-cache enabled, per-packet ACK
is no longer needed. Pure streaming at 921600 baud:

Before (per-packet ACK): 16MB in 9 min (30 KB/s)
After (streaming):       16MB in 3.5 min (77 KB/s)

Selfupdate also switched to streaming mode.
Verified: 16MB CRC32 match on hi3516ev300.

Co-authored-by: Dmitry Ilyin <widgetii@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: widgetii

agent/main.c

@@ -227,9 +227,9 @@ static void handle_write(const uint8_t *data, uint32_t len) {
             uint32_t chunk = pkt_len - 2;
             for (uint32_t i = 0; i < chunk && received < size; i++)
                 dest[received++] = pkt[2 + i];
-            /* Backpressure: COBS-framed ACK after each DATA packet.
-             * Host waits for this before sending next packet. */
-            proto_send_ack(ACK_OK);
+            /* No per-packet ACK — streaming mode. COBS CRC32 per packet
+             * catches any corruption. D-cache makes processing fast
+             * enough to keep up with 921600 baud. */
         } else if (cmd == 0) {
             uint8_t err[5];
             err[0] = ACK_FLASH_ERROR;
@@ -333,8 +333,6 @@ static void handle_flash_write(const uint8_t *data, uint32_t len) {
             uint32_t chunk = pkt_len - 2;
             for (uint32_t i = 0; i < chunk && received < size; i++)
                 staging[received++] = pkt[2 + i];
-            /* Backpressure ACK */
-            proto_send_ack(ACK_OK);
         } else if (cmd == 0) {
             uint8_t err[5];
             err[0] = ACK_FLASH_ERROR;
@@ -540,8 +538,7 @@ static void handle_selfupdate(const uint8_t *data, uint32_t len) {
             uint32_t chunk = pkt_len - 2;
             for (uint32_t i = 0; i < chunk && received < size; i++)
                 dest[received++] = pkt[2 + i];
-            /* Backpressure ACK — host must wait before sending next */
-            proto_send_ack(ACK_OK);
+            /* Streaming mode — no per-packet ACK */
         } else if (cmd == 0) {
             proto_send_ack(ACK_FLASH_ERROR);
             return;

src/defib/agent/client.py

@@ -323,9 +323,9 @@ async def _write_block(self, addr: int, data: bytes) -> bool:
         if cmd != RSP_ACK or resp[0] != ACK_OK:
             return False
 
-        # Send data packets with backpressure: agent sends a COBS-framed
-        # ACK after each DATA packet. Host waits for it before sending
-        # next. Guarantees zero data loss at any speed.
+        # Stream all DATA packets without per-packet ACK.
+        # With COBS bug fixed + D-cache, the agent processes fast
+        # enough to keep up at 921600 baud.
         offset = 0
         seq = 0
         while offset < len(data):
@@ -335,11 +335,6 @@ async def _write_block(self, addr: int, data: bytes) -> bool:
             offset += chunk
             seq += 1
 
-            # Wait for per-packet ACK
-            cmd, resp = await recv_response(self._transport, timeout=10.0)
-            if cmd != RSP_ACK or resp[0] != ACK_OK:
-                return False
-
         # Final CRC verification ACK — may take seconds for large transfers
         crc_timeout = max(60.0, len(data) / 50000)  # ~20µs/byte for CRC32
         cmd, resp = await recv_response(self._transport, timeout=crc_timeout)
@@ -603,7 +598,7 @@ async def selfupdate(
         if cmd != RSP_ACK or resp[0] != ACK_OK:
             return False
 
-        # Send data with per-packet backpressure
+        # Stream all data packets
         offset = 0
         seq = 0
         while offset < len(firmware):
@@ -613,12 +608,6 @@ async def selfupdate(
             offset += chunk
             seq += 1
 
-            # Wait for per-packet ACK
-            cmd, resp = await recv_response(self._transport, timeout=10.0)
-            if cmd != RSP_ACK or resp[0] != ACK_OK:
-                logger.error("Selfupdate packet %d failed: 0x%02x", seq, resp[0])
-                return False
-
         # Wait for CRC verification ACK (agent verifies before jumping)
         cmd, resp = await recv_response(self._transport, timeout=30.0)
         if cmd != RSP_ACK or resp[0] != ACK_OK: