Add LlmIntentClassifier and chat-to-proposal integration tests

[Chris0Jeky]

Parent

Part of #570 (Chat-to-proposal NLP gap)

Problem

The LlmIntentClassifier has zero dedicated test coverage. The chat-to-proposal flow through ChatService → classifier → ParseInstructionAsync is not tested end-to-end with natural language inputs.

Proposed Tests

LlmIntentClassifierTests (new file)

Unit tests documenting current behavior and gaps:

// Current behavior (should pass)
[Theory]
[InlineData("create card 'test'", true, "card.create")]
[InlineData("move card to done", true, "card.move")]
[InlineData("what's the weather?", false, null)]
public void Classify_CurrentPatterns(string input, bool expected, string? intent)

// Known gaps (documenting limitations — expected to fail until #571 is fixed)
[Theory]
[InlineData("can you create new onboarding tasks for non-technical people?")]
[InlineData("I need three new cards for the sprint")]
[InlineData("set up a project board for Q2 planning")]
public void Classify_NaturalLanguage_CurrentlyMisses(string input)

ChatService proposal flow tests

Integration tests verifying the end-to-end path with mock provider:

// Natural language → classifier misses → no proposal attempt (current behavior)
[Fact]
public async Task SendMessage_NaturalLanguage_NoProposalWhenClassifierMisses()

// Explicit RequestProposal → parser fails on natural language → graceful error
[Fact]
public async Task SendMessage_RequestProposal_GracefulErrorOnNaturalLanguage()

// Structured syntax → classifier hits → parser succeeds → proposal created
[Fact]
public async Task SendMessage_StructuredSyntax_ProposalCreated()

Affected Files

• backend/tests/Taskdeck.Application.Tests/Services/LlmIntentClassifierTests.cs (new)
• backend/tests/Taskdeck.Application.Tests/Services/ChatServiceTests.cs (additions)
• backend/tests/Taskdeck.Application.Tests/Services/AutomationPlannerServiceTests.cs (additions)

Acceptance Criteria

• LlmIntentClassifierTests covers all current patterns + documents known gaps
• ChatService tests verify the proposal flow end-to-end
• All tests pass with current codebase (gap tests document limitations, not assert failures)
• Tests serve as regression guardrails for future NLP improvements