Add the SEPTIC motion profile

[Jepson2k]

This PR adds SEPTIC, a seventh-order point-to-point profile whose velocity, acceleration and jerk are all zero at both ends. QUINTIC starts and stops with a jerk step, while SEPTIC ramps jerk in and out. par6 moves with the same profile under the same rule, so both backends offer the same set of profiles.

• Point-to-point only. SEPTIC times a straight joint run, just as RUCKIG does. A cartesian path or a blended chain is timed by TOPPRA instead, as in par6. Timing a path with one polynomial lets its steepest row decide the duration of the whole move: from standby, a 40 mm MoveL planned 24 s.
• Duration. Each joint's displacement against its velocity, acceleration and jerk limits gives a bound, and the duration is the largest of these. A requested duration can stretch it but never shorten it.
• Shared code. SEPTIC and QUINTIC's straight-run case use one per-joint polynomial builder (_septic_samples beside _quintic_samples). QUINTIC's output is unchanged.
• Profile names. VALID_PROFILES and the invalid-profile remedy are now derived from ProfileType, instead of being lists maintained by hand.
• CI lint fix. The Lint check failed under ty 0.0.86 on the async client's robot setter, a line this PR didn't otherwise touch. The setter now takes any waldoctl Robot, as the interface declares, and asserts that it is parol6's own.

Tests

• Limits and jerk ramp. test_septic_holds_its_limits_and_starts_and_stops_without_a_jerk_step uses a move short enough that the jerk limit sets its duration. Without the jerk term, jerk reaches 1.97× the limit. The test also checks that SEPTIC's jerk ramps in and out while QUINTIC's steps, using a threshold that holds at any control rate.
• Cartesian timing. test_a_septic_cartesian_move_is_timed_along_its_path previews a 40 mm MoveL from standby under SEPTIC and under TOPPRA. Before the point-to-point rule, SEPTIC planned 23.76 s against TOPPRA's 2.02 s.
• Velocity test. test_trajectory_respects_joint_velocity_limits now differentiates at the control tick. It used to divide by duration / (N − 1), which understated velocity.
• Results. Locally, 433 tests passed and 8 were skipped. The one failure, test_msg_mounts_on_the_flange_not_inside_the_wrist, is caused by rtree missing from the local venv; the dev extra installs it.

Not in this PR

• _enforce_segment_limits never checks jerk, and its acceleration stretch only takes effect on segments already at full speed. As a result, QUINTIC, LINEAR and TOPPRA along a path that bends can exceed the hard acceleration and jerk limits. This is pre-existing and deserves its own PR.

🤖 Generated with Claude Code

https://claude.ai/code/session_016uVshJDKTz7nCqrMGVxVJU