fix: weight oversample_thick/polar/omega per-contribution, not on the running sum

[gcmullen]

Summary

In the per-pixel accumulation, oversample_thick/polar/omega multiply the running
accumulator I by capture_fraction/polar/omega_pixel on every innermost iteration
(:3026-3028). Since I already holds all prior contributions, each factor
re-multiplies earlier contributions on every later iteration — it compounds
across the subpixel/source/phi/mosaic/thick loops.

Effect

• oversample_thick: sum ratio vs the per-contribution result grows 2.3x -> 42x as
  detector_thicksteps goes 2 -> 25 (layer 1's contribution ends up scaled by
  cf_1cf_2...*cf_N).
• oversample_omega: corr collapses to 0.34 under divergence + oversample.

Fix

Fold each factor into a per-contribution term before accumulating:

double I_contribution = F_cell*F_cell*F_latt*F_latt;
if(oversample_thick) I_contribution *= capture_fraction;
if(oversample_polar) I_contribution *= polar;
if(oversample_omega) I_contribution *= omega_pixel;
I += I_contribution;

Verification

• Default (non-oversample) path: byte-identical (all flags off => I += I_contribution
  is identical to the old I += ...).
• oversample_thick: sum ratio flat ~1.0 across thicksteps 2->25 (was 2.3->42x).
• oversample_polar/omega: parity restored (omega 0.34 -> 1.0).

Scope

nanoBragg.c, +6/-4, one hunk. No API/CLI change. Proposal for your review.