fix: add xhigh effort typing and retain CLI stderr for exit errors

[oubeichen]

Summary

Add xhigh to the Python SDK effort typing and retain CLI stderr so
non-zero subprocess exits can surface the underlying CLI error text in
ProcessError.

Background

Current Claude Code CLI versions (v2.1.112+) support xhigh as an effort level, but
the Python SDK typing still omitted it from:

• ClaudeAgentOptions.effort
• AgentDefinition.effort

That left the typed SDK surface out of sync with current CLI behavior.

Separately, when the CLI failed during startup or argument validation,
the SDK raised a generic ProcessError with:

Error output: Check stderr output for details

This made CLI-side failures harder to diagnose than necessary.

Changes

Add xhigh to the shared effort typing

Introduce a shared effort level type that includes xhigh, and use it for:

• ClaudeAgentOptions.effort
• AgentDefinition.effort

This keeps the typing aligned with current Claude Code CLI support and
avoids duplicating the allowed literal set in multiple places.

Retain CLI stderr for exit errors

Keep the existing live stderr forwarding behavior in _handle_stderr()
for callbacks and debug-to-stderr, and always pipe stderr so
ProcessError can include the CLI error text when the subprocess exits
with a non-zero status.

When no stderr is available, the SDK still falls back to the previous
generic message so existing caller expectations remain unchanged.

Tests

Add regression coverage for:

• ClaudeAgentOptions(effort="xhigh")
• AgentDefinition(effort="xhigh")
• CLI command construction for --effort xhigh
• stderr propagation into ProcessError on subprocess failure

Verified with:

./.venv/bin/python -m pytest tests/test_types.py tests/test_transport.py tests/test_errors.py -q
./.venv/bin/python -m ruff check src/claude_agent_sdk/types.py src/claude_agent_sdk/_internal/transport/subprocess_cli.py tests/test_types.py tests/test_transport.py


Refs #834.