Fix stale initial angle propagation in terminal bending muscles

[Copilot]

• Inspect current muscle angle fix and failing test
• Investigate timing difference between AutoTriggerInterval 20 and 100
• Identify a smaller engine-side fix
• Replace broad muscle-processing fix with minimal constructor-side guard
• Format modified files
• Run targeted/order-dependent validation
• Run full core validation

[chrxh]

@copilot explain your fix

[Copilot]

@copilot explain your fix

The fix in e704fd5 narrows the previous broad muscle-side workaround to the actual construction case. continueConstructionOnBranch was restoring a bending host muscle’s initialAngle via lastObject even when lastObject was not the host muscle’s pivot connection, which could rewrite the wrong angle and produce the 120/240 geometry. Now it still resets the host muscle front angle, but only calls restoreInitialAngleFromPrevious(hostObject, lastObject, ...) when hostObject->connections[0].object == lastObject, i.e. when lastObject is the real pivot. The AutoTriggerInterval=20 case worked because construction advanced before the stale restore affected the final geometry; interval 100 aligned with head/front-angle updates and exposed it.