Fix halved potential weight in the Metropolis changePath action

[RichardHoekstra]

The (1/2.0) factor in changePath()'s ΔE multiplies the whole bracket, including the VTrial(s, xt) and V(s, j) potential terms. But V()/VTrial() already carry their own 1/2 (the harmonic potential returns x²/2), so the potential enters the discretized action at half its intended weight.

The effect is physical, not cosmetic: the sampler draws from an effective oscillator with ω² = 1/2, whose harmonic ground state is 1/√2 ≈ 0.707 instead of the exact 0.5.

Fix: scope the 1/2 to the two kinetic links only, and add the potential difference (VTrial - V) at full weight — in all three (j == end / j == 0 / bulk) branches, so the kinetic and potential parts of the discretized Euclidean action carry their correct relative weight:

ΔE = (1/2)[ (Δx_new)² − (Δx_old)² ] / Δτ²  +  [ V(xt) − V(x_j) ]

Verification: against an independent reimplementation of the same Metropolis update, the corrected rule converges to the harmonic E₀ = 0.5 within Monte Carlo error across several time-step and seed choices, and tightens toward 0.5 under Δτ refinement. The original rule sits at 0.707 and stays biased as Δτ → 0, i.e. it is a genuine physical bias rather than a discretization artifact.

Scope kept minimal — only the ΔE weighting is touched.