feat(joint-follow): host-side entry for CMD_JOINT_FOLLOW (0x08)

[luomaofu]

Summary

Adds the host-side entry point for CMD_JOINT_FOLLOW (0x08) and pins its payload
layout against the firmware, so a caller can drive the slave arm's follow loop
without the SDK re-deriving the frame by hand.

The command is the one the isomorphic teleop stack uses on the slave. Its payload
carries four groups — target position, target velocity, kp, kd — and no tau;
the firmware computes the gravity feedforward and the limit wall itself, which is
what makes the frame round-trip cheaper than the older send_mit_all path. Nothing
in the payload is self-describing, so a drifted layout is silent until the arm
misbehaves on hardware. The new test closes that gap.

No consumer breaks. The command is additive: an arm without it answers ERR with
cmd = 0x08, which the existing capability check already treats as "not supported".

Changes

• src/litearm/_protocol.py — register CMD_JOINT_FOLLOW = 0x08 alongside the
  other command codes.
• src/litearm/arm.py — Arm.joint_follow(q, dq, kp, kd) sends the frame. The
  four groups are packed in firmware order and the call is a single round trip.
• src/litearm/errors.py — map the command's error code so a rejected frame
  reports which command was refused rather than an anonymous failure.
• tests/test_protocol_sync.py — extract the field order from the firmware header
  usb_cmd.h with a regex, then send a real frame with a distinct sentinel in every
  slot and read the bytes back segment by segment. A reordering or a dropped field
  fails the test rather than shipping.
• tests/test_error_codes.py, tests/test_full_coverage.py — extend the existing
  tables to cover the new command code.

Testing

$ python -m pytest tests -q
.....ss................................................................. [ 76%]
........................................................................ [ 87%]
........................................................................ [ 98%]
........                                                                 [100%]
654 passed, 2 skipped in 60.92s (0:01:00)

The two skips are the live-hardware cases in tests/test_live.py, gated on
LITEARM_LIVE=1; they are skipped by design on a machine without an arm attached,
not a pre-existing failure.

The payload-layout test was mutation-checked both ways: reordering two segments in
the expected order, and dropping one, each turn it red.

Issues

None. No issue was opened for this work, so there is nothing to close.

[github-actions]

🎉 This PR is included in version 1.1.0 🎉

The release is available on GitHub release

Your semantic-release bot 📦🚀