Non utf8 output causes panic in interpreter discovery

On some machines (as detected by Zed crash telemetry), toolchain discovery panics because a probed executable’s stdout is not valid UTF-8.

For example, we received:

b"\xCE\xC4\xBC\xFE\xB2\xBB\xB4\xE6\xD4\xDA: E:\\<redacted>\\-c\r\n"

This (according to my LLM) is GBK-encoding of 文件不存在: E:\<redacted>\-c, “File does not exist.”

I'm not sure why the executable appears to interpret -c as a filename, but I don't think we should panic on its output. Using String::from_utf8_lossy instead of String::from_utf8(...).unwrap() should address this.

在解释器发现代码中搜索使用 String::from_utf8(...).unwrap() 处理探测到的可执行文件输出的代码。更新输出处理逻辑，使非 UTF-8 的 stdout 不会触发 panic，然后验证当探测程序输出无效 UTF-8 时发现过程仍能完成。