Unexpected/unusable behavior with steering prompts and subagents with 1.0.49 version. Worked fine in 1.0.48.

### Describe the bug

I use parallel subagents a lot and with the latest version the experience has degraded a lot and has become unusable.
1. The subagents keep on running indefinitely mostly. And even if I ask the main agent to stop them or ask them to consolidate their findings they don't stop. They keep on running in loops. 1-2 hrs and even more.
2. Steering prompts also doesn't work since the turns of those agents never completes. Its like the main agent doesnt pick up new prompts from the queue as it keeps on waiting for the subagents and subagents never complete. With 1.0.48, steering was very good, it stopped current work (of main agent) and I could send important directions if I saw it deviating. With new version the steered prompts are just added to the queue and are picked up very late.
3. Explore agents work fine, general-purpose aren't working.

Evidences

### Affected version

1.0.49

### Steps to reproduce the behavior

_No response_

### Expected behavior

_No response_

### Additional context

_No response_

Reproduce the behavior in version 1.0.49 with parallel general-purpose subagents and steering prompts, then compare it with 1.0.48. Trace why general-purpose subagents do not terminate and why queued steering prompts are delayed; done means subagents stop or consolidate promptly and steering interrupts the current work as expected.

由索引模型根据 Issue 内容生成。