Support 1-d array data in profile exporter

[nv-hwoo]

1. Support 1-d array data in profile exporter so that when the input/output involves more than one data value (or array of values), then it captures all the data, rather than just the first element. This is an incremental step towards supporting full N-dimensional tensor in profile exporter. Currently, we don't need to support N-dimensional tensors because we are mostly interested in the text string or token ids (which is just a 1-d array). Additionally, N-dimensional tensor support requires more changes to PA since response outputs doesn't store shape information of the tensors returned from the model so we need to retrieve that information somewhere.
2. Support other data types such as uint, fp, and etc.

Example:

When given trtllm engine the following input:

{
    "token_ids": [1, 2, 3, 4, 5, 6, 7, 8],
    "input_length": [8],
    "request_output_len": [10],
    ...
}

what we got from PA before the change is

{
    "request_inputs": {
        "token_ids": 1,  // missing remaining values
        "input_length": 8,
        "request_output_len": 10,
    },
    ...
}

After the fix, we get the full array of token ids (same for any other fields in the profile export json file):

{
    "request_inputs": {
        "token_ids": [1, 2, 3, 4, 5, 6, 7, 8],
        "input_length": 8,
        "request_output_len": 10,
    },
    ...
}