fix: honour flip in DHLink.A() when the joint ET is not last (#563)

[youdie006]

Fixes #563.

Problem

DHLink.A() decided whether to negate the joint variable by checking self.ets[-1].isflip. For a link with a nonzero a, d or alpha, _to_ets() places the joint ET (Rz(flip=flip)) before the constant ETs (tz(d) / tx(a) / Rx(alpha)), so self.ets[-1] is a constant ET whose isflip is always False. As a result flip=True was silently ignored for such links and q was not negated, so DHRobot.fkine disagreed with the link's own ETS.

import roboticstoolbox as rp, math
L = rp.RevoluteDH(a=1.0, flip=True)
L.A(0.5).A[1, 0]        # returned +0.47943 (== +sin(0.5)), wrong
L.ets.eval([0.5])[1, 0] # -0.47943 (== sin(-0.5)), correct

A control link with a = 0 (where the joint ET is last) already behaved correctly, which pinpoints the cause.

Fix

Check self.isflip (the joint ET's flip, via Link.isflip -> self.v.isflip) instead of the last ET. When the joint ET is last (a = d = alpha = 0) the two are equal, so previously-correct links are unaffected. One line in DHLink.A().

Testing

• Added test_flip_with_nonzero_a in tests/test_DHRobot.py: asserts RevoluteDH(a=1.0, flip=True).A(q).A equals the link's own ets.eval([q]) and that A[1, 0] == sin(-q).
• Verified red -> green: the new test fails on the current code (+0.479 vs desired -0.479) and passes with the fix. Isolated the change against the full tests/test_DHRobot.py: baseline had 2 failures (this new test + the pre-existing test_payload drift), and after the one-line fix only test_payload remains -- exactly this test flips fail -> pass, no other test changes. Sanity-checked that a=0 flip stays correct and non-flip links are unchanged.

Thanks to @yucchengyi for the report and the precise root-cause diagnosis.

---

This change was prepared with AI assistance and reviewed by me before submission.