avoid redundant parsing of repeated value in RleDecoder

[jp0317]

Which issue does this PR close?

No issue

Rationale for this change

performance improvement

What changes are included in this PR?

Moving the parsing of repeated_value out of the loop

Are there any user-facing changes?

Not really: although it adds Clone restriction to the get_batch<T>(), the T in practice always implements Clone.

[tustvold]

Have you run any benchmarks for this?

[jp0317]

Have you run any benchmarks for this?

not really. I was doing some experiments on decoding dictionary indices and there's noticeable portion of cpu time spent on parsing this repeated value. And it seems obvious that the repeated value can be parsed outside the loop, avoiding the redundant parsing. With the change the cpu time spent inside this get_batch got reduced in my experiments.