datastore: `ipfs repo stat` seems to hang with large repos on Windows

#### Version information:
0.4.5-dev, 56058c1 with HAMT sharding merged in (unrelated)

#### Type: bug

#### Priority: P2

#### Description:

On a repo with 312695 objects, on a HDD formatted with NTFS, `ipfs repo stat` seems to take forever to finish.
I assume it's doing something inefficient when checking for filesizes, as `ipfs refs local` finishes in 18 seconds.

```
C:\Users\matin>sh -c "time ipfs repo stat"

Error: request canceled

real 12m41.638s
user 0m0.000s
sys 0m0.031s
```
```
C:\Users\matin>sh -c "time ipfs refs local | wc"
312695 312695 14728215

real 0m18.022s
user 0m0.375s
sys 0m0.952s
```
(on a side note, `refs local` also seems slightly slow, but that might just be an IO bound)