test(kernel-language-model-service): Add queue language model service for testing (#710)

This PR adds a queue language model service to the monorepo. The
implementation is intended for mocking language model behavior where a
remote API or local model execution are both unavailable. The intended
streaming response is pushed to the model prior to sampling.

Also adds tests to kernel-test which exercise the basic shape of a
LanguageModelService.

Author: grypez

packages/kernel-language-model-service/package.json

@@ -33,6 +33,16 @@
         "default": "./dist/ollama/nodejs.cjs"
       }
     },
+    "./test-utils": {
+      "import": {
+        "types": "./dist/test-utils/index.d.mts",
+        "default": "./dist/test-utils/index.mjs"
+      },
+      "require": {
+        "types": "./dist/test-utils/index.d.cts",
+        "default": "./dist/test-utils/index.cjs"
+      }
+    },
     "./package.json": "./package.json"
   },
   "files": [

packages/kernel-language-model-service/src/test-utils/index.ts

@@ -0,0 +1,4 @@
+export { makeQueueService } from './queue/service.ts';
+export { makeQueueModel } from './queue/model.ts';
+export type { QueueLanguageModel } from './queue/model.ts';
+export type { QueueLanguageModelService } from './queue/service.ts';

packages/kernel-language-model-service/src/test-utils/queue/README.md

@@ -0,0 +1,49 @@
+# Queue-based Language Model Service (Testing Utility)
+
+[`makeQueueService`](./service.ts) is a testing utility that creates a `LanguageModelService` implementation for use in tests. It provides a queue-based language model where responses are manually queued using the `push()` method and consumed by `sample()` calls.
+
+## Usage
+
+1. Create a service using `makeQueueService()`
+2. Create a model instance using `makeInstance()`
+3. Queue responses using `push()` on the model instance
+4. Consume responses by calling `sample()`
+
+Note that `makeInstance` and `sample` ignore their arguments, but expect them nonetheless.
+
+## Examples
+
+### Basic Example
+
+```typescript
+import { makeQueueService } from '@ocap/kernel-language-model-service/test-utils';
+
+const service = makeQueueService();
+const model = await service.makeInstance({ model: 'test' });
+
+// Queue a response
+model.push('Hello, world!');
+
+// Consume the response
+const result = await model.sample({ prompt: 'Say hello' });
+for await (const chunk of result.stream) {
+  console.log(chunk.response); // 'Hello, world!'
+}
+```
+
+### Multiple Queued Responses
+
+```typescript
+const service = makeQueueService();
+const model = await service.makeInstance({ model: 'test' });
+
+// Queue multiple responses
+model.push('First response');
+model.push('Second response');
+
+// Each sample() call consumes the next queued response
+const first = await model.sample({ prompt: 'test' });
+const second = await model.sample({ prompt: 'test' });
+
+// Process streams...
+```