refactor(vscode): drop chat-completion fallback in chat textarea autocomplete

Both supported autocomplete models (Codestral, Mercury) expose FIM, and
AutocompleteModel.supportsFim() unconditionally returns true. The non-FIM
branch in ChatTextAreaAutocomplete together with the getChatSystemPrompt
/ getChatUserPrompt helpers was never reached.

Also removes the accompanying orphaned spec that imported paths that no
longer exist (core/config/ProviderSettingsManager, api/transform/stream).

packages/kilo-vscode/src/services/autocomplete/chat-autocomplete/ChatTextAreaAutocomplete.ts

@@ -98,22 +98,9 @@ export class ChatTextAreaAutocomplete {
     let response = ""
 
     try {
-      // Use FIM if supported, otherwise fall back to chat-based completion
-      if (this.model.supportsFim()) {
-        await this.model.generateFimResponse(prefix, suffix, (chunk) => {
-          response += chunk
-        })
-      } else {
-        // Fall back to chat-based completion for models without FIM support
-        const systemPrompt = this.getChatSystemPrompt()
-        const userPrompt = this.getChatUserPrompt(prefix)
-
-        await this.model.generateResponse(systemPrompt, userPrompt, (chunk) => {
-          if (chunk.type === "text") {
-            response += chunk.text
-          }
-        })
-      }
+      await this.model.generateFimResponse(prefix, suffix, (chunk) => {
+        response += chunk
+      })
 
       const latencyMs = Date.now() - startTime
 
@@ -153,35 +140,6 @@ export class ChatTextAreaAutocomplete {
     }
   }
 
-  /**
-   * Get system prompt for chat-based completion
-   */
-  private getChatSystemPrompt(): string {
-    return `You are an intelligent chat completion assistant. Your task is to complete the user's message naturally based on the provided context.
-
-## RULES
-- Provide a natural, conversational completion
-- Be concise - typically 1-15 words
-- Match the user's tone and style
-- Use context from visible code if relevant
-- NEVER repeat what the user already typed
-- NEVER start with comments (//, /*, #)
-- If the user is in the middle of typing a word (e.g., "hel"), include the COMPLETE word in your response (e.g., "hello world" not just "lo world")
-- This allows proper prefix matching to remove the overlap correctly
-- Return ONLY the completion text, no explanations or formatting`
-  }
-
-  /**
-   * Get user prompt for chat-based completion
-   */
-  private getChatUserPrompt(prefix: string): string {
-    return `${prefix}
-
-TASK: Complete the user's message naturally. 
-- If the user is mid-word (e.g., typed "hel"), return the COMPLETE word (e.g., "hello world") so prefix matching can work correctly
-- Return ONLY the completion text (what comes next), no explanations.`
-  }
-
   private async buildPrefix(userText: string, visibleCodeContext?: VisibleCodeContext): Promise<string> {
     return buildChatPrefix(userText, visibleCodeContext?.editors)
   }

packages/kilo-vscode/src/services/autocomplete/chat-autocomplete/__tests__/ChatTextAreaAutocomplete.spec.ts

@@ -1,84 +0,0 @@
-import { ChatTextAreaAutocomplete } from "../ChatTextAreaAutocomplete"
-import { ProviderSettingsManager } from "../../../../core/config/ProviderSettingsManager"
-import { AutocompleteModel } from "../../AutocompleteModel"
-import { ApiStreamChunk } from "../../../../api/transform/stream"
-
-describe("ChatTextAreaAutocomplete", () => {
-  let autocomplete: ChatTextAreaAutocomplete
-  let mockProviderSettingsManager: ProviderSettingsManager
-
-  beforeEach(() => {
-    mockProviderSettingsManager = {} as ProviderSettingsManager
-    autocomplete = new ChatTextAreaAutocomplete(mockProviderSettingsManager)
-  })
-
-  describe("getCompletion", () => {
-    it("should work with non-FIM models using chat-based completion", async () => {
-      // Setup: Model without FIM support (like Mistral)
-      const mockModel = new AutocompleteModel()
-
-      vi.spyOn(mockModel, "hasValidCredentials").mockReturnValue(true)
-      vi.spyOn(mockModel, "supportsFim").mockReturnValue(false)
-      vi.spyOn(mockModel, "generateResponse").mockImplementation(async (systemPrompt, userPrompt, onChunk) => {
-        // Simulate streaming chat response
-        const chunks: ApiStreamChunk[] = [{ type: "text", text: "write a function" }]
-        for (const chunk of chunks) {
-          onChunk(chunk)
-        }
-        return {
-          cost: 0,
-          inputTokens: 15,
-          outputTokens: 8,
-          cacheWriteTokens: 0,
-          cacheReadTokens: 0,
-        }
-      })
-
-      // @ts-expect-error - accessing private property for test
-      autocomplete.model = mockModel
-
-      const result = await autocomplete.getCompletion("How to ")
-
-      expect(mockModel.generateResponse).toHaveBeenCalled()
-      expect(result.suggestion).toBe("write a function")
-    })
-  })
-
-  describe("cleanSuggestion", () => {
-    it("should filter code patterns (comments, preprocessor)", () => {
-      // Comments - filtered by the regex check in cleanSuggestion
-      expect(autocomplete.cleanSuggestion("// comment", "")).toBe("")
-      expect(autocomplete.cleanSuggestion("/* comment", "")).toBe("")
-      expect(autocomplete.cleanSuggestion("* something", "")).toBe("")
-
-      // Code patterns
-      expect(autocomplete.cleanSuggestion("#include", "")).toBe("")
-      expect(autocomplete.cleanSuggestion("# Header", "")).toBe("")
-    })
-
-    it("should filter empty content", () => {
-      // Empty content is filtered by postprocessAutocompleteSuggestion
-      expect(autocomplete.cleanSuggestion("", "")).toBe("")
-    })
-
-    it("should accept natural language suggestions", () => {
-      expect(autocomplete.cleanSuggestion("Hello world", "")).toBe("Hello world")
-      expect(autocomplete.cleanSuggestion("Can you help me", "")).toBe("Can you help me")
-      expect(autocomplete.cleanSuggestion("test123", "")).toBe("test123")
-      expect(autocomplete.cleanSuggestion("What's up?", "")).toBe("What's up?")
-    })
-
-    it("should accept symbols in middle of text", () => {
-      expect(autocomplete.cleanSuggestion("Text with # in middle", "")).toBe("Text with # in middle")
-      expect(autocomplete.cleanSuggestion("Hello // but not a comment", "")).toBe("Hello // but not a comment")
-    })
-
-    it("should truncate at first newline", () => {
-      expect(autocomplete.cleanSuggestion("First line\nSecond line", "")).toBe("First line")
-    })
-
-    it("should remove prefix overlap", () => {
-      expect(autocomplete.cleanSuggestion("Hello world", "Hello ")).toBe("world")
-    })
-  })
-})