future.retry() retries futures regardless of its status

**What happened**:
I was checking my logs and noticed that some futures were executed more than once.
It appears that future.retry() will unconditionally re-schedule* the future.

* I'm not sure if re-scheduling is the proper expression.

**What you expected to happen**:
I expected that future.retry() would re-scheduled ONLY futures that failed,
i.e.: that had a status == "lost" or "error".
But it appears that it re-schedules futures even if they are 'pending'
That is what I understood from the documentation:

https://distributed.dask.org/en/latest/api.html#distributed.Client.retry

retry(futures, asynchronous=None)[source]
Retry failed futures

**Minimal Complete Verifiable Example**:
In the code below, I expected to ready "started" once; when the future was started.
But it is printed infinitely many times and the task never terminates.

```python
import time

import distributed

def ughz():
print("started!")
time.sleep(2)

def main():
cluster = distributed.LocalCluster(threads_per_worker=1, n_workers=1)
client = distributed.Client(cluster)
f = client.submit(func=ughz, pure=False)
while f.status != "finished":
print(f.status)
print()
f.retry()

if __name__ == '__main__':
main()
```

**Anything else we need to know?**:
It was hard to find which values a future.status can assume.
I had to check client.py source code.

**Environment**:

- Dask version: 2020.12.0
- Python version: 3.8.5
- Operating System: CentOS Linux release 7.2.1511 (Core)
- Install method (conda, pip, source): conda