Skip psutil, rely on explicit render job count

The default core count is lower than practically every CPU in usage.
This is optimal to avoid eating up more framebuffers than CPU L3 cache
and slowing down all render jobs.

corrscope/corrscope.py

@@ -10,11 +10,9 @@ from multiprocessing.shared_memory import SharedMemory
 from pathlib import Path
 from queue import Queue, Empty
 from threading import Thread
-from typing import Iterator, Optional, List, Callable, Tuple, Dict, Union, Any
+from typing import Iterator, Optional, List, Callable, Dict, Union, Any
 
 import attr
-import numpy as np
-import psutil
 
 from corrscope import outputs as outputs_
 from corrscope.channel import Channel, ChannelConfig, DefaultLabel
@@ -25,7 +23,6 @@ from corrscope.renderer import (
     Renderer,
     RendererConfig,
     RendererParams,
-    RendererFrontend,
     RenderInput,
 )
 from corrscope.settings.global_prefs import Parallelism
@@ -715,17 +712,7 @@ class CorrScope:
 
         parallelism = self.arg.parallelism
         if parallelism and parallelism.parallel:
-            # determine thread count
-            def _nthread():
-                # Spawn one process per physical core (not hyper-thread).
-                #
-                # On a Ryzen 5 5600X, the render processes are over twice as slow (
-                # and have substantially higher jitter) when we spawn one per
-                # hyper-thread rather than one per core.
-                ncores = psutil.cpu_count(logical=False)
-                return min(ncores, parallelism.max_render_cores)
-
-            nthread = _nthread()
+            nthread = parallelism.max_render_cores
 
             with self._load_outputs(nthread):
                 play_parallel(nthread)