perf: combine whole fine storage and batched stream completion

Integrate only the two reviewed improvements onto eec4677. Storage tiles use
one whole wide fine allocation; step(n) completes its unchanged one-stream
schedule with one wait per call. Raw exchange retains the final interval;
an open exchange window spans the completed batch.

Storage-source: ab699854ae2fb0c994a14bfd7124dddbe50d8a88
Original-sync-author: Claude <claudeworker@noreply.invalid>
Sync-source: ad1d82cb7f
Sync-doc-correction: 512dcc923b
Validated-local-sync-source: 65c001c3294a6eb45283568bf4967c673b33b19a
Sync-stable-patch-id: a825dcd47b33c5c4d9be31e0071d2306eb344d0b
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

tests/validation/run_refined_dfg.py

@@ -17,7 +17,7 @@ HERE = Path(__file__).resolve().parent
 DEFAULT_SOURCE = HERE.parents[1]
 BOUNDS = (40, 24, 320, 140)
 BASE_BLOCK = (2, 881)  # Exact 166 x 881 domain; no geometric padding.
-FINE_BLOCK = (8, 8)  # 240 x 568 including four fine halo cells.
+FINE_BLOCK = (240, 568)  # Whole fine allocation, including its refinement ghost band.
 SPEED = 0.01