Spatial × FEC — spatial diversity keeps corruption localized (sim + measure)

[josephnef]

Part of the spatial-diversity roadmap #127. Depends on #B, #C.

Goal

Establish and measure the strongest cross-layer link in the roadmap: spatial
diversity keeps corruption localized, which is the precondition the fused-FEC
sub-block-salvage layer needs to work.

The argument

The fused-FEC stack (docs/fused-fec.md) only gains when corruption is localized
within a frame. Deep fades push corruption into the frame-wide regime where
sub-block salvage collapses (measured survivor fraction ~0.36; "neither inter-frame
scheme recovers"). Spatial diversity is precisely the tool that kills deep fades —
so it does not merely lower the corrupt-frame rate, it moves corruption out of the
frame-wide regime and into the localized regime salvage exploits. On RX, combining
before the soft demapper also yields cleaner soft information → better localization.

Scope

• Extend the fused-FEC simulation to model per-chain fading + combining and show
  the shift in the survivor-fraction distribution (frame-wide → localized).
• Measure on-air: sub-block survival and delivered-block gain with vs without
  spatial diversity, at matched corrupt-frame rate.
• Quantify how diversity + soft-decode compose (both localize; are they additive?).

[josephnef]

Sim done → PR #147. Diversity cuts SNR variance (deep fades are variance), moving frames out of the frame-wide-loss bin into the SBI-salvageable one — improving SBI's precondition. Modeled at matched mean SNR (isolating the variance effect): static/correlated barely helps (lost 0.31 vs single 0.33), mobile/decorrelated collapses it (0.12) and lifts salvageable-of-corrupt 0.59→0.86. Corruption localization is a mobility effect.