.NET: [Feature Branch] Add Azure Functions hosting support for durable workflows

[kshyju]

Motivation and Context

This PR adds Azure Functions hosting support for durable workflows, building on the foundational workflow APIs introduced in #3648. While #3648 added console/hosted-service support for running workflows as Durable Task orchestrations, this PR enables workflows to be hosted as Azure Functions with automatic trigger registration, removing the need for manual function definitions.

Note: This PR targets the feat/durable_task feature branch.

Key scenarios enabled:

• Hosting durable workflows as Azure Functions with zero boilerplate trigger definitions
• Automatic HTTP trigger registration for starting workflow orchestrations via HTTP
• Automatic activity trigger registration for workflow executors
• Automatic entity trigger registration for AI agent executors within workflows
• Deduplication of shared executors across multiple workflows

Description

This PR introduces Azure Functions hosting infrastructure for durable workflows, along with new public APIs and shared utilities.

New Public APIs

Two new extension methods are added to FunctionsApplicationBuilderExtensions in the Microsoft.Agents.AI.Hosting.AzureFunctions namespace. These complement the existing ConfigureDurableAgents method.

API	Description
ConfigureDurableWorkflows	Configures the Azure Functions app to host durable workflows. Automatically registers orchestrations, generates function triggers (HTTP, activity, entity) for each workflow and its executors, and sets up the middleware pipeline. This is the recommended entry point for workflow-only applications.
ConfigureDurableOptions	Configures the Azure Functions app with full durable options, including both agents and workflows. Use this when your application needs to host both durable agents and durable workflows in the same function app.

public static FunctionsApplicationBuilder ConfigureDurableWorkflows(
    this FunctionsApplicationBuilder builder,
    Action<DurableWorkflowOptions> configure);

public static FunctionsApplicationBuilder ConfigureDurableOptions(
    this FunctionsApplicationBuilder builder,
    Action<DurableOptions> configure);

Note: All other new types added in this PR (FunctionMetadataFactory, DurableWorkflowsFunctionMetadataTransformer, BuiltInFunctionExecutor, etc.) are internal and not part of the public API surface.

Usage Examples

Sequential Workflow hosted in Azure Functions (Sample: 01_SequentialWorkflow)

OrderLookup orderLookup = new();
OrderCancel orderCancel = new();
SendEmail sendEmail = new();

Workflow cancelOrder = new WorkflowBuilder(orderLookup)
    .WithName("CancelOrder")
    .AddEdge(orderLookup, orderCancel)
    .AddEdge(orderCancel, sendEmail)
    .Build();

var builder = FunctionsApplication.CreateBuilder(args);

// Register the workflow — HTTP and activity triggers are generated automatically.
builder.ConfigureDurableWorkflows(workflows =>
{
    workflows.AddWorkflow(cancelOrder);
});

builder.Build().Run();

Once running, the workflow can be started via HTTP:

curl -X POST http://localhost:7071/api/workflows/CancelOrder/run \
  -H "Content-Type: text/plain" \
  -d "ORDER-123"

Fan-out/Fan-in Workflow with AI Agents (Sample: 02_ConcurrentWorkflow)

ParseQuestionExecutor parseQuestion = new();
AIAgent physicist = chatClient.AsAIAgent("You are a physics expert.", "Physicist");
AIAgent chemist = chatClient.AsAIAgent("You are a chemistry expert.", "Chemist");
AggregatorExecutor aggregator = new();

Workflow workflow = new WorkflowBuilder(parseQuestion)
    .WithName("ExpertReview")
    .AddFanOutEdge(parseQuestion, [physicist, chemist])
    .AddFanInEdge([physicist, chemist], aggregator)
    .Build();

builder.ConfigureDurableWorkflows(workflows =>
{
    workflows.AddWorkflow(workflow);
});

Using ConfigureDurableOptions for mixed agent + workflow scenarios:

builder.ConfigureDurableOptions(options =>
{
    options.Agents.AddAgent("MyAgent", agentFactory);
    options.Workflows.AddWorkflow(myWorkflow);
});

Internal Infrastructure Changes

Component	Description
DurableWorkflowsFunctionMetadataTransformer	Dynamically registers HTTP, activity, and entity triggers for each workflow and its executors, with deduplication for shared executors
FunctionMetadataFactory	Shared factory for creating entity, HTTP, and activity trigger metadata — extracted from DurableAgentFunctionMetadataTransformer to eliminate duplication
BuiltInFunctions	Added RunWorkflowOrchestrationHttpTriggerAsync and InvokeWorkflowActivityAsync with input validation
BuiltInFunctionExecutor	Extended to dispatch workflow HTTP and activity entry points
ServiceCollectionExtensions	Made AddDurableTaskWorker conditional; moved client-dependent registrations inside if (clientBuilder is not null) block
DurableWorkflowRunner	Constructor changed to public for DI activation (class remains internal)

Package Updates

Package	Old Version	New Version
Microsoft.DurableTask.Client	1.18.0	1.19.1
Microsoft.DurableTask.Client.AzureManaged	1.18.0	1.19.0
Microsoft.DurableTask.Worker	1.18.0	1.19.0
Microsoft.DurableTask.Worker.AzureManaged	1.18.0	1.19.0
Microsoft.Azure.Functions.Worker.Extensions.DurableTask	1.11.0	1.13.1

Workaround

Suppressed AD0001 in AzureFunctions csproj — temporary workaround for Microsoft.DurableTask.Analyzers v0.2.0 bug (ArgumentNullException on node). To be removed when a fixed analyzer version is available.

Validation/Testing

Samples Added:

Sample	Description
samples/Durable/Workflow/AzureFunctions/01_SequentialWorkflow	Demonstrates sequential workflow with shared executors across two workflows (CancelOrder, OrderStatus). No Azure OpenAI required.
samples/Durable/Workflow/AzureFunctions/02_ConcurrentWorkflow	Demonstrates fan-out/fan-in workflow with AI agent executors. Requires Azure OpenAI.

Integration Tests: Integration tests have been added (WorkflowSamplesValidation.cs) to run and validate both Azure Functions workflow samples, ensuring end-to-end functionality works as expected.

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the Contribution Guidelines
• All unit tests pass, and I have added new tests where possible
• Is this a breaking change? No

[cgillum]

Overall looks good. Added a few comments, mostly non-blocking.