[Optimization] Tunes badger for raftwal

[harshil-goel]

We observed that when running dgraph at scale, things started to get slow. The throughput in diskless mode went from 8k to less than 1k within minutes for continous mutations. We noticed that most(80%) of time was being taken while PreFetch() in badger for raft-wal. This PR fixes that. Now the throughput stabalizes around 7k expect for when compaction happens.

---

This change is 

[manishrjain]

Please add a big comment. Also, prefix the title with [Optimization].

Reviewed 1 of 1 files at r1.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @harshil-goel)

---

raftwal/storage.go, line 526 at r1 (raw file):

		iopt := badger.DefaultIteratorOptions
		iopt.PrefetchValues = false

Add a big comment here that: multiple things: One that in LSM tree mode, which is how WAL is opened, there's no need to prefetch values, because the values are colocated with the keys. And secondly, if we set this to false, we no longer see the spikes in mutation throughput exhibited by Dgraph.

[martinmr]

Just one thing. From the description I gather this was affecting diskless mode. What about the normal mode? From the code it looks like it's also being affected by this change. If that's the case it's a good idea to mention it in the commit description and/or comment.

Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @harshil-goel)