[FEA] Support Retry on HostColumnarToGpu

[revans2]

Is your feature request related to a problem? Please describe.
HostColumnarToGpu is concatenating the data in host memory before sending it down to the GPU. We cannot really support split and retry here without having a way for our HostColumnarBuilder to also split it's input. Plus that feels like a very badly configured query where we don't have enough memory to process a single batch of data that matches the target batch size.

Describe the solution you'd like
This needs to be a little bit different from the other AbstractGpuCoalesceIterator retry implementations. Here we just want the retry to be in HostToGpuCoalesceIterator.concatAllAndPutOnGPU All of the rest of the processing is done on the CPU.