Standardize null pyworker load/perf on 100

Using 1 confused the serverless capacity math. Set workload_calculator,
benchmark target throughput, and client cost all to 100 — the conventional
default the rest of the system expects.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

workers/null/client.py

@@ -36,7 +36,7 @@ async def reserve(
     start = time.monotonic()
     log.info("[%s] POST /reserve duration=%ss", label, duration)
     try:
-        resp = await endpoint.request("/reserve", payload, cost=1)
+        resp = await endpoint.request("/reserve", payload, cost=100)
         elapsed = time.monotonic() - start
         log.info("[%s] returned after %.1fs: %s", label, elapsed, resp.get("response"))
         return resp["response"]

workers/null/worker.py

@@ -117,11 +117,11 @@ async def reserve_worker(**params: object) -> dict:
 
     if params.get(BENCHMARK_SENTINEL):
         # The framework computes max_throughput = workload / time during the
-        # startup benchmark. A null worker has no throughput concept (a
+        # startup benchmark. A null worker has no real throughput concept,
-        # reservation is a unitless slot), so we deliberately take ~1s with
+        # so we deliberately take ~1s with workload=100 to pin
-        # workload=1 to pin max_throughput to ~1.0. Without this the
+        # max_throughput to ~100. Without this the near-instant benchmark
-        # near-instant benchmark would report hundreds of thousands of
+        # would report hundreds of thousands of workload/sec, distorting
-        # workload/sec, distorting any downstream capacity math.
+        # any downstream capacity math.
         await asyncio.sleep(1.0)
         return {"ok": True, "benchmark": True}
 
@@ -173,7 +173,7 @@ worker_config = WorkerConfig(
             # it to a free worker (or spins up a new one).
             max_queue_time=0.0,
             remote_function=reserve_worker,
-            workload_calculator=lambda _payload: 1.0,
+            workload_calculator=lambda _payload: 100.0,
             benchmark_config=BenchmarkConfig(
                 generator=lambda: {BENCHMARK_SENTINEL: True},
                 runs=1,