Second-order power weighting in plotSpectra/plotCDF (#383)

[gustavdelius]

Closes #383. Refines the plotting placement of #382 for the power argument of
the plotSpectra…/plotCDF… families.

Stacked on #396 (#382). This PR's base is the second-order-plot-382
branch, so the diff shows only the #383 changes. Merge #396 first, then this
retargets cleanly onto master.

The problem

plotSpectra() forms the plotted value with the weight and the location both at
the node: N_j · w_j^power at w = w_j. For a cell average N_j the
transformed density N(w) w^power is represented at the geometric bin centre
w*_j = w_j sqrt(β) by N_j (w*_j)^power, so the node is doubly off:

• the location error — the half-bin shift of Second-order plotting: locate bin-averaged values at geometric bin centres + Array representation tag #382;
• the w^power weighting error — O(power·Δx), largest for the most common
  power = 2 (Sheldon/biomass-per-log-size) plot, where w_j^power undershoots
  (w*_j)^power by a full factor β.

Fix

• plot_spectra() (density): under second_order_w[["bin_average"]],
  evaluate both the w^power weight and the plotted location at the bin
  centre w* = w sqrt(β) (species, resource, total, background), reusing
  bin_midpoints() from Second-order plotting: locate bin-averaged values at geometric bin centres + Array representation tag #382. The default auto wlim follows the plotted grid
  so the centre shift does not clip the end bins.
• animate.MizerSim() gets the same treatment;
  plotlySpectra()/plotlyCDF() inherit it (thin do.call wrappers);
  plotSpectraRelative() is the special case where the w^power factor
  cancels in 2(N2−N1)/(N1+N2), so only its x-location shifts — which it
  inherits automatically.
• plotCDF() is the opposite case: a CDF value is cumulative up to a size,
  a boundary quantity, so plot_cdf() maps the density back to the bin
  edges (w = w*/sqrt(β)) — keeping the cumulative on the nodes — while the
  increments stay centre-weighted, so each value·Δw_k = N_k (w*_k)^power Δw_k
  is the second-order bin integral of N w^power.

Explicitly out of scope

getFlux() also has a power argument but is a flux at the boundary → a
point value at the face, correctly weighted at the node; it is left untouched.

Tests

test-plots.R: for power = 0/1/2 the spectrum x shifts by sqrt(β) and the
value scales by β^(power/2) (weight evaluated at the centre); the default path
keeps x on the nodes; plotCDF x stays on the node/edge grid (never the
centres) and is monotonic; plotSpectraRelative shifts x to centres but its
value is power-independent (cancellation). Default vdiffr snapshots unchanged.
Full suite: 2842 pass.

Docs

NEWS entry; the numerical-details vignette's plotting note extended to cover the
w^power-at-centre weighting and the CDF's edge placement.

🤖 Generated with Claude Code