Possible DB-API fetchall() performance improvements

Currently the cursor implementation for [fetchall](https://github.com/dropbox/PyHive/blob/65076bbc8697a423b438dc03e928a6dff86fd2cb/pyhive/common.py#L129) indirectly iterates over the result set via `fetchone` rather than directly iterating over the result set similar to prestodb's python [DB-API](https://github.com/prestodb/presto-python-client/blob/master/prestodb/dbapi.py#L306).

I'm not certain whether it's viable to use this approach, however I mocked up a basic example which shows that the proposed solution is about 10x faster.

```python
import timeit

def current():
class Cursor:
def __init__(self):
self._iterator = iter(range(10000))

def fetchone(self):
try:
return next(self._iterator)
except StopIteration:
return None

def fetchall(self):
return list(iter(self.fetchone, None))

cursor = Cursor()
cursor.fetchall()

def proposed():
class Cursor:
def __init__(self):
self._iterator = iter(range(10000))

def fetchall(self):
return list(self._iterator)

cursor = Cursor()
cursor.fetchall()

print(timeit.timeit("current()", number=1000, setup="from __main__ import current"))
print(timeit.timeit("proposed()", number=1000, setup="from __main__ import proposed"))
```
resulted in 2.999s and 0.205s respectively, i.e., the proposed solution is about 10x faster.