MCP semantic_code_search: Update handling for no code embeddings

What does this MR do and why?

As part of &19655, we are going to implement ad-hoc embeddings indexing of projects. With ad-hoc indexing, we are not immediately indexing Code Embeddings for all eligible projects. Instead, we will perform Code Embeddings indexing for a project on the first time a semantic code search is triggered on the project.

In this MR, we update the semantic code search process to handle the different "no code embeddings" scenarios:

Note: "no code embeddings" mean that the project has no Ai::ActiveContext::Code::Repository OR its state != ready.

1. If Ai::ActiveContext::Code::Repository record does not exist for a project:
   • trigger the AdHocIndexingWorker (introduced in #577330 (closed))
   • return a message indicating that the embeddings initial indexing has just started
2. If Ai::ActiveContext::Code::Repository exists with state=failed
   • return a message indicating that the embeddings indexing failed
3. If Ai::ActiveContext::Code::Repository exists with state NOT ready or failed
   • return a message indicating that the embeddings initial indexing is still ongoing

References

• Related Issue: [Code Embeddings Query] Trigger ad-hoc indexing... (#576633 - closed)
• Issue introducing the AdHocIndexingWorker: [Code Embeddings] Create AdHocIndexingWorker (#577330 - closed)

Screenshots or screen recordings

Scenario: project has code embeddings

Scenario: project has NO code embeddings

Screen_Recording_2025-10-29_at_17.18.57

Scenario: project embeddings indexing failed

Scenario: project embeddings initial indexing is in progress

How to set up and validate locally

Prerequisites

1. Setup MCP server on your GDK
2. Setup the Code Embeddings Indexing pipeline on your GDK

Testing

Test the semantic_code_search tool using MCP Inspector or your preferred client for MCP testing.

1. For a project with code embeddings, the tool should return results successfully

   Setup note: you can set up this project by following Prerequisites Step 2

2. For a project with NO code embeddings but eligible for it:

   • an Ai::ActiveContext::Code::Repository record should be created for the project
   • the record should have a state of code_indexing_in_progress or embedding_indexing_in_progress (the record is initially created with state=pending, but since the ad-hoc indexing is kicked off, the indexing process should have started by the time you are checking

   Setup note: this should be any project under the gitlab-duo namespace without an Ai::ActiveContext::Code::Repository record

3. For a project with Ai::ActiveContext::Code::Repository record with state=failed

   • the tool should error with a message indicating that the indexing failed

   Setup note: you can set up this project by following Prerequisites Step 2, then updating the project's Ai::ActiveContext::Code::Repository record to state=failed

4. For a project with Ai::ActiveContext::Code::Repository record with state NOT ready or failed

   • the tool should error with a message indicating that initial indexing is still ongoing

   Setup note: you can set up this project by following Prerequisites Step 2, then updating the project's Ai::ActiveContext::Code::Repository record to code_indexing_in_progress=failed

MR acceptance checklist

Evaluate this MR against the MR acceptance checklist. It helps you analyze changes to reduce risks in quality, performance, reliability, security, and maintainability.

Related to #576633 (closed)