feat: support stderr logging in Jupyter notebook environments

[BabyChrist666]

Summary

Fixes #156

In Jupyter notebook environments, passing sys.stderr directly to subprocess doesn't work as expected - the stderr output is lost and users have no visibility into server errors or crashes. This makes debugging extremely difficult as the notebook cell just hangs with no feedback.

Changes

• Added _is_jupyter_environment() function to detect Jupyter/IPython notebook environments
• In Jupyter mode, stderr is captured as a pipe and read asynchronously
• Stderr output is printed with red ANSI color for visibility in notebooks
• In non-Jupyter environments, behavior remains completely unchanged (direct stderr passthrough)

How It Works

1. On client initialization, we detect if running in Jupyter/IPython
2. If in Jupyter, we set capture_stderr=True which pipes stderr instead of redirecting
3. An async stderr_reader task reads the pipe and prints to the notebook output
4. Red ANSI coloring makes errors stand out in the notebook

Technical Details

• Detection checks for IPKernelApp in config or ZMQInteractiveShell class name
• Falls back gracefully if IPython is not installed
• Windows compatibility maintained via FallbackProcess stderr wrapping

Test plan

• Added TestJupyterStderrSupport test class with 4 tests
• Test Jupyter detection returns False in normal Python
• Test graceful handling when IPython not installed
• Test stderr capture doesn't block the client
• Test continuous stderr output is handled properly
• All existing stdio tests pass

🤖 Generated with Claude Code