docs(register): reframe C-37/C-38 as near-term preconditions for probabilistic reconciliation

[Polichinel]

Append-only register notes (counts unchanged, 41/24/17) reflecting this session's cross-repo reframe of the reconciliation effort.

Why

• FAO (rusty_bucket) needs no reconciliation — pure-grid ensemble aggregated up, sums by construction.
• The imminent UN-agency deliverable is an FAO-like grid ensemble that does reconcile against a CM model, with ~1024 pooled draws → probabilistic reconciliation, not pgm_cm_point.
• The views-postprocessing reconciler is already probabilistic-ready (vectorized over samples, bit-exact at S=1000) — so the open work is in the consumer (pipeline-core) plus two preconditions:

Changes

• C-37 (sample-alignment): reframed from "deferred" to an active design gate — does the deliverable reconcile grid draws to a country point total (no alignment) or to country draws (needs defined alignment; CM models are point-only/independent today)? Gates pipeline-core#200.
• C-38 (memory residual): the chunk-by-time obligation is now near-term — to be honored and dry-run-measured by pipeline-core#200's PredictionFrameEnsembleManager consumer.

No reconciler code change. Companion to the cross-repo issue updates (vpp #31/#39/#40/#41/#45, pipeline-core #194/#200, views-reporting #72/#149, views-models #143/#77/#144).

🤖 Generated with Claude Code