fix: preserve thinking=disabled as --thinking-disabled flag

[shubham108prakash-hub]

Fixes #693

Problem
When thinking={"type": "disabled"} was set in ClaudeAgentOptions,
the SDK was converting it to --max-thinking-tokens 0.

This loses the user's explicit intent because:

• --max-thinking-tokens 0 = ambiguous, may be ignored
• thinking={"type": "disabled"} = explicit opt-out

Anthropic-compatible third-party providers treat these differently,
causing thinking to remain enabled even when the user explicitly disabled it.

Fix
Changed the disabled case in _build_command() to append
--thinking-disabled flag instead of setting tokens to 0.

Before:
elif t["type"] == "disabled":
resolved_max_thinking_tokens = 0

After:
elif t["type"] == "disabled":
cmd.append("--thinking-disabled")

Testing

• Added test_build_command_with_thinking_disabled in tests/test_transport.py
• Verifies --thinking-disabled is present in command
• Verifies --max-thinking-tokens is NOT present when disabled
• All 38 tests passing