[R-package] Sort indices in lgb.cv() (fixes #2503)

[jameslamb]

This PR attempts to fix the issue reported by @rgranvil in #2503 . You can confirm by building the R package from this branch and running the sample code provided in that issue.

I effectively just mirrored what was done on the Python side in #2510 .

[StrikerRUS]

Sorry, don't have R environment to confirm the fix.

[Laurae2]

It seems to work correctly here.

[guolinke]

@jameslamb did you check these ?

LightGBM/R-package/R/lgb.cv.R

Lines 235 to 239 in b4bb38d

	dtrain <- slice(data, seq_len(nrow(data))[-folds[[k]]])
	setinfo(dtrain, "weight", getinfo(data, "weight")[-folds[[k]]])
	setinfo(dtrain, "init_score", getinfo(data, "init_score")[-folds[[k]]])
	setinfo(dtest, "weight", getinfo(data, "weight")[folds[[k]]])
	setinfo(dtest, "init_score", getinfo(data, "init_score")[folds[[k]]])

for the indices in weight and init_score.

[guolinke]

ping @jameslamb @Laurae2 again

[jameslamb]

@guolinke I did not. Looking...

[guolinke]

@jameslamb
I am afraid that the index orders used in feature and (weight, label, ...) are different.

[StrikerRUS]

@jameslamb Any updates?

[jameslamb]

@StrikerRUS @guolinke sorry for the delay! I've opened #2572 to track this, so we can have the discussion in an open issue instead of in a comment on a closed PR.

I have a PR to fix this coming momentarily.

[StrikerRUS]

@jameslamb Thanks! I was worrying that this issue can be lost in the closed PR.