ClaudeSDKClient.interrupt() is ignored silently when called immediately after query()

[vadhavka-deshaw]

Bug description:
Calling interrupt() immediately (less than roughly 2 seconds) after query() has no observable effect. Claude continues generating a full response as if interrupt() was never called. The issue disappears if there is a slight delay between the two calls, suggesting a race condition.

Steps to reproduce:

import time
from claude_agent_sdk import ClaudeAgentOptions, ClaudeSDKClient
from claude_agent_sdk.types import ResultMessage, StreamEvent

options = ClaudeAgentOptions(
    cli_path="/path/to/claude",
    include_partial_messages=True,
)

ts = lambda: time.strftime("%H:%M:%S")

async with ClaudeSDKClient(options=options) as client:
    await client.query("Write a 300 word essay on machine learning")
    print(f"[{ts()}] query sent")
    await client.interrupt()
    print(f"[{ts()}] interrupt acknowledged")
    t0 = time.monotonic()

    chars_after = 0
    async for msg in client.receive_response():
        if isinstance(msg, StreamEvent):
            d = msg.event.get("delta", {})
            chars_after += len(d.get("text", "") + d.get("thinking", ""))
        elif isinstance(msg, ResultMessage):
            dt = time.monotonic() - t0
            verdict = "BUG" if chars_after else "OK"
            print(
                f"{verdict}: {chars_after} chars in {dt:.1f}s after interrupt (stop_reason={msg.stop_reason})"
            )
            break

The output shows:

[15:20:49] query sent
[15:20:49] interrupt acknowledged
BUG: 2464 chars in 23.6s after interrupt (stop_reason=end_turn)

Even after interrupt() resolves without errors, Claude continues to generate the full response. The stop_reason is end_turn, not a signal indicating early termination. Adding a time.sleep() of a few seconds before the interrupt() call causes the interrupt to work correctly.