Support remote inference on Triton Inference Server with ease of use

[MMelQin]

This pull request adds support for performing inference on remote Triton Inference Server, with ease of use

• updated the TritonModel class to fully support and encapsulate the Triton client of the specific model hosted in Triton Inference Server. This is done by parsing the key elements of Triton model config.pbtxt file (actual model files not required), and dynamically setting the input and output parameters using the parsed model configurations. User app is relived from setting any arguments except for the server network location, and in future versions, TLS keys for secure communication.
• enhanced the NamedModel to include the support of Triton model repository (which may have multiple folders each for a specific model), making it possible to support hybrid inference scenario, i.e. local in-proc hosted as well as Triton hosted models, all transparent to the user app.
• updated the app context class to support the setting for Triton server network location, as well as calling the instantiated TritonModel to connect to the remote server.
• Added and tested example app showcasing remote inference on Triton Inference Server.

Also added the following files to the example application folder:

• Example Triton Model repo for the client side, i.e. with only model folder and its config.pbtxt file.
• Example Shell script to set the required env vars, though if these are not set, the command line option can be used instead.

The Quality Gate failure is due to the example app has duplicated code, which is true as it has exactly the same app inference operator as the Spleen Seg app with in-proc Torch model hosting. We have each example app folder contain all necessary code for ease of use, hence have to tolerate the duplicates with a few hundred lines of code.

[mocsharp]

LGTM. Thank you.

[sonarqubecloud]

Quality Gate failed

Failed conditions
10.2% Duplication on New Code (required ≤ 3%)

See analysis details on SonarQube Cloud