.Net: Pass Kernel to connectors and implement automatic OpenAI function calling - #3754

Closes #3687
Closes #2932

This PR includes three commits:

Commit 1: Simply updates XML comments for Kernel arguments to standardize them.

Commit 2: Updates the connector interfaces (e.g. IChatCompletion) to accept an optional Kernel argument. It's passed through everywhere, but the actual connector implementations largely ignore it, since they don't need anything from it right now.

Commit 3: Updates the {Azure} OpenAI connector to support automatic function calling, whether those functions are explicitly specified in OpenAIPromptExecutionSettings or are implicitly available via the passed in Kernel. The behavior is opt-in, via a revised setting on the OpenAIPromptExecutionSettings class. Rather than just specifying a string (which had been "auto", "none" or a specific function name), this now exposes a property of a new type FunctionCallBehavior, which can be set to various instances to achieve the desired results:

• FunctionCallBehavior.EnableKernelFunctions - the functions in the provided kernel will be sent but won't be automatically invoked if requested
• FunctionCallBehavior.AutoInvokeKernelFunctions - the functions in the provided kernel will be sent and will be automatically invoked if requested
• FunctionCallBehavior.EnableFunctions(functionsList, autoInvoke) - the functions specified will be sent and whether they'll be automatically invoked is determined by the autoInvoke argument and by whether they're available in the provided kernel
• FunctionCallBehavior.RequireFunction(function, autoInvoke) - the specified function will be sent to the service as a request to function call it

The auto-invoke behavior is wired up based on whether autoInvoke is true. The FunctionCallBehavior.AutoInvokeKernelFunctions can also be opted-into in a service-agnostic manner by setting a named property in the base PromptExecutionSettings class's dictionary; future service implementations can respect that as well if they either have a built-in notion of function calling (which some other LLMs do) or if we want to simulate it with a planner baked into the connector implementation.

Please review this both for my API choices and for implementation details. Roger Barreto (@rogerbarreto), in particular I'd appreciate it if you could look at how streaming is handled. Gina Triolo (@gitri-ms), I'd appreciate it if you could review the function calling APIs and implementation support.

With all of this, this is now the "Hello, world" of function calling:

Kernel kernel = new KernelBuilder()
    .WithOpenAIChatCompletion("gpt-3.5-turbo-1106", apiKey)
    .ConfigurePlugins(plugins => plugins.AddPluginFromObject<TimePlugin>())
    .Build();

OpenAIPromptExecutionSettings settings = new() { FunctionCallBehavior = FunctionCallBehavior.AutoInvokeKernelFunctions };
Console.WriteLine(await kernel.InvokePromptAsync("What is the current time?", settings));