**kwargs unexpectedly parses flags for later chained method

Suppose we have the following program

```python
from dataclasses import dataclass

@dataclass
class MyClass:
x: int

def transform(self, **kwargs):
return MyClass(self.x * kwargs.get("multiplier", 2))

def do_something(self, msg):
print(f"{msg}: {self.x}")

my_obj = MyClass(3)
```

And we call it with `python -m fire my_module my_obj transform --multiplier=3 do_something --msg="test"`, then the `--msg` is parsed when calling `transform` into the `kwargs` dict. Then it errors saying `--msg` is missing. Can we instead support parsing flags before the next chained method call? The kwargs is useful for example, to transform the object in ways that only implementation classes know what to do.

My current workarounds are (1) pass msg as positional arg to `do_something`, (2) have `do_something` take `**kwargs` and call `transform` within it. Neither is ideal as (1) would surprise anyone who uses my code and (2) adds extra plumbing that could be avoided.

使用所示的 `python -m fire` 命令以及鏈式呼叫 `transform` 和 `do_something` 重現此行為。首先追蹤鏈式方法呼叫的引數解析；完成的標準是，`transform` 之後的 flags 會傳遞給 `do_something`，而不是被吸收到 `transform` 的 `kwargs` 中，並為此範例提供 coverage。

由索引模型根據 Issue 內容生成。