P9 complete: compiled kernels validated (byte-identical, 78-80 GSA samples/min/core), scaling report

Graviton3 c7g.8xlarge, 360 GSE165845 arrays: 360/360 GTCs identical between the Rust
kernels and the pure-Python paths; 1 worker 78-80 samples/min (pure Python 29.6).
Scaling 1->8 workers efficiency 0.87-0.90; above that the 125 MB/s gp3 disk caps the
run (~620 samples/min) and, on a RAM disk, a fixed per-worker start-up cost limits a
360-sample batch to efficiency 0.48 at 32 workers (K-025). Worker state is pickled once.
docs/validation/P9.md, result JSONs, guide/README/CHANGELOG updates.

src/simurg/pipeline/genotype.py

@@ -9,6 +9,8 @@ import json
 import math
 import multiprocessing as mp
 import os
+import pickle
+import tempfile
 from concurrent.futures import ProcessPoolExecutor
 from dataclasses import dataclass, field, replace
 from pathlib import Path
@@ -147,6 +149,13 @@ def _init(
     )
 
 
+def _init_from_file(path: str) -> None:
+    """Worker initializer: the shared state is pickled once by the parent (not once per worker, which serialised
+    worker start-up: ~0.25 s per worker for a GSA manifest context) and read by every worker in parallel."""
+    with open(path, "rb") as fh:
+        _init(*pickle.load(fh))
+
+
 def _process(job: Job) -> dict[str, Any]:
     ctx: ManifestContext = _W["ctx"]
     params: GenotypeParams = _W["params"]
@@ -338,13 +347,17 @@ def run_genotype(opts: GenotypeOptions, rep: Reporter) -> int:
             _init(ctx, params, auto_1_22, probes, started, beads, th.bg_offset)
             results = [_process(j) for j in jobs]
         else:
-            with ProcessPoolExecutor(
-                workers,
-                mp_context=mp.get_context("spawn"),
-                initializer=_init,
-                initargs=(ctx, params, auto_1_22, probes, started, beads, th.bg_offset),
-            ) as ex:
-                results = list(ex.map(_process, jobs))
+            opts.out.mkdir(parents=True, exist_ok=True)
+            fd, state = tempfile.mkstemp(prefix=".simurg-workers-", suffix=".pkl", dir=opts.out)
+            try:
+                with os.fdopen(fd, "wb") as fh:
+                    pickle.dump((ctx, params, auto_1_22, probes, started, beads, th.bg_offset), fh, protocol=5)
+                with ProcessPoolExecutor(
+                    workers, mp_context=mp.get_context("spawn"), initializer=_init_from_file, initargs=(state,)
+                ) as ex:
+                    results = list(ex.map(_process, jobs))
+            finally:
+                Path(state).unlink(missing_ok=True)
     except KeyboardInterrupt:
         rm.status = "interrupted"
         rm.write(opts.out / "run_manifest.json")