Retry the first handshake after USB re-enumeration #10

Context

Arm.connect() (src/litearm/arm.py:883) builds the session and then performs one handshake:

# 握手: 固件版本
self._raw_write(P.CMD_GET_FIRMWARE)
try:
    _, verp = self._a.expect(P.RSP_FIRMWARE, 1.5, "get_firmware",
                             echo_cmd=P.CMD_GET_FIRMWARE)
    self.firmware = verp.decode(errors="replace").strip()
except Exception:
    self.close()
    raise

One write, one 1.5 s window, then close() and re-raise. There is no retry anywhere in the function.

Expected

After a USB re-enumeration the first connect() succeeds, the same as any other connect().

Actual

After the board re-enumerates — re-plugging the cable, or re-attaching USB passthrough to a VM — the first handshake goes unanswered and connect() raises. Calling it again a second or two later always works.

Measured on hardware: across four re-plugs in one session, the first connection failed every time and every retry succeeded. It never happened on a link that had stayed up.

The cost is not the retry itself, it is the diagnosis. The failure presents as a firmware that is not answering, so the natural response is to check the cable, the permissions and the firmware — none of which is the problem. connect() has already established that the port opened and the reader thread is running; what is missing is only that the firmware needed a moment after enumeration.

Reproduce

On a board where the USB link has just been re-established:

arm = litearm.Arm()
try:
    arm.connect()          # raises: get_firmware 无应答
except litearm.MotionTimeoutError:
    time.sleep(1.5)
    arm.connect()          # succeeds

I did not run this against the Python implementation; it is how the C++ port behaves, and the C++ port kept this function's structure on purpose so the two stay comparable.

Suggested direction

A short bounded retry of the handshake inside connect() — the port is already open at that point, so the retry is cheap and has no side effects. The sub-connect() behaviour is otherwise correct: FirmwareMismatchError and the version checks should not be retried, only the missing reply.