fix: use sys.executable instead of hardcoded "python" in tests

Tests that launch the API server via subprocess used a hardcoded
"python" command which fails on systems where only "python3" is
available (e.g. macOS). Using sys.executable ensures the same
interpreter running pytest is used for subprocesses.

Fixes 36 test errors on macOS environments.

atroposlib/tests/api_test_utils.py

@@ -1,4 +1,5 @@
 import subprocess
+import sys
 import time
 
 import requests
@@ -16,7 +17,7 @@ def launch_api_for_testing(max_wait_for_api: int = 10) -> subprocess.Popen:
     # Use subprocess instead of multiprocessing to avoid inheriting pytest args
     api_proc = subprocess.Popen(
         [
-            "python",
+            sys.executable,
             "-m",
             "atroposlib.cli.run_api",
             "--host",

atroposlib/tests/test_api_compression.py

@@ -5,6 +5,7 @@ import json
 import os
 import signal
 import subprocess
+import sys
 import time
 
 import pytest
@@ -27,7 +28,7 @@ def wait_for_api_server(max_wait=10):
 def api_server():
     proc = subprocess.Popen(
         [
-            "python",
+            sys.executable,
             "-m",
             "atroposlib.cli.run_api",
             "--host",

atroposlib/tests/test_api_messages_handling.py

@@ -5,6 +5,7 @@ Tests for API server message handling, particularly for SFT (Supervised Fine-Tun
 import os
 import signal
 import subprocess
+import sys
 import time
 
 import pytest
@@ -30,7 +31,7 @@ def api_server():
     # Start the API server as a subprocess
     proc = subprocess.Popen(
         [
-            "python",
+            sys.executable,
             "-m",
             "atroposlib.cli.run_api",
             "--host",