fix: Improve error message when validating IO config

[tanmayv25]

Some of the user models can have large number of model inputs/ouptuts. Without this change the model tensor name is not a part of the error message. This change improves the error logging of the server offering better debugging ability to the user.

There is also a fix included which correctly prints a model input as an input instead of as an output.

The testing is updated to expect this new behavior here: triton-inference-server/server#7340

For the following model configuration:

name: "missing_datatype"
max_batch_size: 4
platform: "tensorflow_savedmodel"
input [
  {
    name: "input"
    dims: [ 2, -1 ]
  }
]
output [
  {
    name: "output"
    data_type: TYPE_FP32
    dims: [ 1 ]
  }
]

Before

E0611 19:35:52.422993 138 model_repository_manager.cc:1371] "Poll failed for model directory 'missing_datatype': model output must specify 'data_type' for missing_datatype"

After

E0611 19:31:47.991606 9486 model_repository_manager.cc:1371] "Poll failed for model directory 'missing_datatype': model input 'input' must specify 'data_type' for missing_datatype"