Fix one-timestep logging skew and document GetTorque() sign convention - #5

Per-step CSV logging in demo_vgoswec.cpp mixed pre-step kinematics with post-step torque/time, skewing every row by one timestep (~0.005 s) and making power_w = -tau*omega incorrect. Both the GUI (ui.IsRunning) and headless loops had identical bugs.

Changes (src/demo_vgoswec.cpp only)

• Bug 1 — logging skew (both loops): Move rsda->GetAngle() / rsda->GetVelocity() to after DoStepDynamics, grouping all four logged reads post-step:

// Before: pitch_rad/pitch_vel read PRE-step → one-sample skew vs. time/torque
const double pitch_rad = rsda->GetAngle();
const double pitch_vel = rsda->GetVelocity();
system.DoStepDynamics(dt);
// ...
const double pto_tau = rsda->GetTorque();  // post-step

// After: all quantities sampled at the same post-step timestamp
system.DoStepDynamics(dt);
// ...
const double pitch_rad = rsda->GetAngle();
const double pitch_vel = rsda->GetVelocity();
const double exc_tau   = exc_provider->GetLatestExcitationTorque();
const double pto_tau   = rsda->GetTorque();

• Bug 2 — sign convention (both loops): Add a comment above rsda->GetTorque() documenting the expected sign per docs/CONTROLLERS.md (positive torque opposes positive θ; P_abs = −τ·ω) and the passive-run validation: pto_torque_nm should carry the opposite sign to flap_pitch_vel_rads, yielding net-positive power_w. No preemptive negation — comment only.