fix(rne): attribute a sandwiched static link's mass to its preceding joint, not the following one

[petercorke]

Problem

Robot.rne()'s link-grouping attached a static (fixed) link to the first
joint encountered scanning forward through the link list. Correct for a
trailing run of static links after the last joint (#636), but wrong for a
static link sandwiched between two joints (joint_A -> static -> joint_B):
it's rigidly welded to joint_A's output, not joint_B's, so its mass was
being misattributed to the wrong joint's torque.

Found while reviewing old third-party issues -- #483 proposed a fix for
exactly this, filed against the pre-#636 code. Verified the underlying
problem is still real on current main (after both #636/#684 and the
separate force-propagation-direction fix in #691) using an independent
virtual-work (finite-difference of CoM height vs. q) ground truth, not just
against rne() itself.

test_invdyn_static2's previous expected values were themselves wrong --
copied from test_invdyn's different topology (where the second link
does rotate with the second joint) rather than derived for its own. So
this bug was being actively protected by a passing test, not caught by one.

Fix

Restructures the grouping to walk each static link's .parent chain up to
its nearest joint ancestor (correct under branching too, not just a flat
list scan). This also makes group[0] unconditionally the joint, which
simplifies the forward/backward recursion's joint lookups and the
kinematic-transform construction that #636's fix had to special-case for a
joint that "might not be the last link in the group."

Rederives test_invdyn_static2's gravity/velocity/inertial-term
assertions for its actual topology (joint2's torque, Coriolis terms, and
inertia-matrix row/column must all be exactly zero -- nothing with mass is
ever downstream of it), and adds a dedicated regression test
(test_invdyn_sandwiched_static_link, +inertia variant) isolating the
exact scenario from #483.

Fixes #483.

Verification

• Red/green: reverted just the Robot.py fix (keeping the corrected
  tests) and confirmed all 3 new/updated tests fail against the old code,
  with exactly the misattributed-torque symptom; restored the fix and
  confirmed they pass.
• Full suite: 744 passed, 0 failed, 82 skipped
  (pytest tests/ --ignore=tests/test_blocks.py)
• test_invdyn_trailing_static_link/_with_inertia (RNE/inertia() silently wrong (not crashing) for URDF robots with fixed/structural joints #636's own regression
  tests) still pass -- confirms the restructuring doesn't reintroduce that
  bug.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 0% with 15 lines in your changes missing coverage. Please review.
✅ Project coverage is 0.00%. Comparing base (a0ef01c) to head (e4fb84a).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/roboticstoolbox/robot/Robot.py	0.00%	15 Missing ⚠️

Additional details and impacted files

@@          Coverage Diff          @@
##            main    #693   +/-   ##
=====================================
  Coverage   0.00%   0.00%           
=====================================
  Files        143     143           
  Lines      14103   14091   -12     
=====================================
+ Misses     14103   14091   -12     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.