Teaches `hfs.ls('gs://bucket/')` to list the files and directories at the top-level of the bucket.

In `main` that command raises because this line of `_ls_no_glob` raises:

```python3
maybe_sb_and_t, maybe_contents = await asyncio.gather(
    self._size_bytes_and_time_modified_or_none(path), ls_as_dir()
)
```

In particular, `statfile` raises a cloud-specific, esoteric error about a malformed URL or empty
object names:

```python3
async def _size_bytes_and_time_modified_or_none(self, path: str) -> Optional[Tuple[int, float]]:
    try:
        # Hadoop semantics: creation time is used if the object has no notion of last modification time.
        file_status = await self.afs.statfile(path)
        return (await file_status.size(), file_status.time_modified().timestamp())
    except FileNotFoundError:
        return None
```

I decided to add a sub-class of `FileNotFoundError` which is self-describing: `IsABucketError`.

I changed most methods to raise that error when given a bucket URL. The two interesting cases:

1. `isdir`. This raises an error but I could also see this returning `True`. A bucket is like a
   directory whose path/name is empty.

2. `isfile`. This returns False but I could also see this raising an error. This just seems
   convenient, we know the bucket is not a file so we should say so.

---

Apparently `hfs.ls` had no current tests because the globbing system doesn't work with Azure
https:// URLs. I fixed it to use `AsyncFSURL.with_new_path_component` which is resilient to Azure
https weirdness. However, I had to change `with_new_path_component` to treat an empty path in a
special way. I wanted this to hold:

```
actual = str(afs.parse_url('gs://bucket').with_new_path_component('bar'))
expected = 'gs://bucket/bar'
assert actual == expected
```

But `with_new_path_component` interacts badly with `GoogleAsyncFSURL.__str__` to return this:

```
'gs://bucket//bar'
```