Batch processing added to Migraphx worker

A major change to the Migraphx worker that adds post-version 1 features.

Implement batch processing in migraphx. In the previous version, migraphx would evaluate each image separately even though they arrived in a batch.
Saving a different compiled version of each model (*.mxr file) for each batch size that is requested. Migraphx includes a batch size in each compiled model, so this is the only way to support different batch sizes without requiring recompilation. File names are now suffixed with batch size; for instance the model resnet50-v2-7.onnx, compiled with batch size of 64, is saved as resnet50-v2-7_b64.mxr
Updated example script to submit requests with the batch and timeout parameters.
Logging of elapsed time for migraphx evaluation, to enable performance testing and comparison by MIGraphX team.

Coding note: the new batch processing required a significant loop refactoring in Migraphx::doRun. Pseudocode for the handling order now looks like this:

For each batch
    Call migraphx::eval(), response is array of arrays
    For each request in batch
        For each image in request
            Add result array to response
        Send response

There are two nested try-catch blocks, since some exceptions apply to the entire batch and some apply only to a single request.

Xilinx / inference-server

Batch processing added to Migraphx worker #52

A major change to the Migraphx worker that adds post-version 1 features.

Coding note: the new batch processing required a significant loop refactoring in Migraphx::doRun. Pseudocode for the handling order now looks like this: