enhancement: Dynamically updating CUDA EP options

[krishung5]

This PR is based on #242 to add the flexibility to allow users to provide CUDA EP options in a flexible fashion. Note that this PR doesn't include the update for TRT options.

Before, the CUDA EP options are set within the parameters field:

...
parameters { key: "cudnn_conv_algo_search" value: { string_value: "0" } }
parameters { key: "gpu_mem_limit" value: { string_value: "4294967200" } }
...

After this enhancement, users are able to set any kinds of CUDA EP options just like the TRT EP:

optimization { execution_accelerators {
  gpu_execution_accelerator : [ {
    name : "cuda"
    parameters { key: "cudnn_conv_use_max_workspace" value: "0" }
    parameters { key: "use_ep_level_unified_stream" value: "1" }}
  ]
}}

For backward compatibility, setting CUDA EP options within the parameters field should still work.

Related PRs:
triton-inference-server/backend#100
triton-inference-server/core#368
Testing: triton-inference-server/server#7328

[krishung5]

Updated doc in commit 6ecda12. Can ignore the commit for formatting.

[gedoensmax]

Thanks a lot for taking a look at my PR. This looks great to me.

[tanmayv25]

@gedoensmax @krishung5 Great work improving the parameter specifications in ORT backend!