fix(test): poll for last stderr line in test_stderr_available_on_error

The poll loop broke as soon as the first stderr line ("Debug: worker
starting") appeared, then asserted the second line ("Error: something
went wrong") was also present. The drain thread reads stderr line-by-line,
so on 3.14 Windows the loop could break after line 1 before line 2 was
captured — a pre-existing race the condition papered over. Poll for the
last-written line so both are guaranteed present. Verified 8/8 on Windows
3.14.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: rustyconover

tests/table_in_out/test_client.py

@@ -259,8 +259,10 @@ def test_stderr_available_on_error(self) -> None:
 
         while elapsed < timeout:
             stderr_output = client.get_worker_stderr()
-            # Check if we have the expected content
-            if "Debug: worker starting" in stderr_output:
+            # Poll for the LAST line written — the drain thread reads stderr
+            # line-by-line, so seeing the first line doesn't guarantee the
+            # second has been captured yet (a race that surfaced on Windows).
+            if "Error: something went wrong" in stderr_output:
                 break
             time.sleep(poll_interval)
             elapsed += poll_interval