Correctness Fix: Block before proposing mutations and improve conflict key generation

[manishrjain]

This PR fixes two things:

• Before proposing mutations, Dgraph didn't block to ensure that it has seen all the committed transaction updates. So, the index calculation had side-effects as shown by count index tests, where a previous update hasn't been registered yet and a new one is made, which caused incorrect count to be generated, hence causing the count index to be wrong.

This PR fixes that by ensuring that Dgraph waits for the Alpha involved to have reached StartTs. Only then would it propose a new update. This ensures that all the updates are in correct order.

• Conflict Keys: Previously, Dgraph was generating conflict keys based on (S, P) for all list values, and (S, P, O) for all uids. This PR changes that by using (S, P) for singular predicates, without considering whether they are uid or value; and using (S, P, O) for list uids/values. This PR also adds conflict keys for indices, treating them as list of uids.

• Based on tests by @danielmai , this PR does not have any impact on the live loader throughput.

---

This change is 

[gitlw]

The tests are failing, I've triggered another run of the tests to rule out transient failures.
Other than that, the PR LGTM.

Reviewable status: 0 of 4 files reviewed, all discussions resolved (waiting on @gitlw and @martinmr)

[gitlw]

Reviewable status: 0 of 4 files reviewed, all discussions resolved (waiting on @gitlw and @martinmr)

[martinmr]

Just one minor comment.

Reviewed 4 of 4 files at r1, 2 of 2 files at r2.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @manishrjain)

---

worker/groups.go, line 203 at r2 (raw file):

	// Propose schema mutation.
	var m pb.Mutations
	// schema for a reserved predicate is not changed once set.

minor: maybe move or remove this comment. It seems out of context now.