[RodStraightSection] Fix rest shape double-offset for non-first sections

[lkarstensen]

Summary

Fixes #223.

RodStraightSection::getRestTransformOnX computed the rest position as x_start + x_used, but x_used is already the global curvilinear abscissa. This placed rest-shape nodes at 2 * x_start + local_offset instead of x_used for any section that is not the first in the wireMaterials list.

Example (stiff section 420 mm + flex section 30 mm):

	x_start	x_used	Old result	Correct
Section 0	0	210	210 ✓	210
Section 1	420	430	850 ✗	430

The ~420 mm offset between rest shape and mechanical DOFs caused AdaptiveBeamForceFieldAndMass to generate huge elastic restoring forces every timestep, leading to simulation explosion on any disturbance (e.g. vessel-wall contact).

Fix

// Before
global_H_local.set(type::Vec3(x_start + x_used, 0.0, 0.0), Quat());

// After
global_H_local.set(type::Vec3(x_used, 0.0, 0.0), Quat());

This matches the pattern already used correctly by RodSpireSection and RodMeshSection. Zero effect on single-section wires (x_start = 0).

[hugtalbot]

Hi @lkarstensen
Thanks a lot for your contribution! SOFA core-devs will review your work
However, could you add a test highlighting the problem that you fixed please? 🙏

[epernod]

Suggested change

	global_H_local.set(type::Vec3(x_used, 0.0, 0.0), Quat());
	global_H_local.set(type::Vec3(x_used - x_start, 0.0, 0.0), Quat());

Thank you for your contribution, indeed there is an error here the x_start should not be added to the x_used which is already the global curvilinear abscissa. However for me the correct fix is to deduce the x_start as the sections are used in reverse order.
If your tool is composed of 2 straightSection:
section 1: length 400
section 2: length 200

when tool will goes out the method will be called on the section 2 with the values:
x_used -> equal to full length -> 600,
x_start = 400 (length of section 1), therefor the transform will be computed at relative value of 600 - 400 = 200 the last position of this section