fix(vector_io): wire file_processors provider into vector store file insertion

[alinaryan]

When a file is posted to a vector store via POST /v1/vector_stores/{id}/files, the mixin code checks for a file_processor_api to process the file (e.g., using pypdf or docling for PDF
parsing). However, that attribute was never wired up — no vector_io provider received the file_processors dependency from the resolver. So vector store file insertion always fell
through to the legacy PyPDF chunking path, regardless of which file_processors provider was configured.

This PR fixes that by:

• Adding Api.file_processors to optional_api_dependencies for all vector_io provider specs
• Threading the dependency through each provider's factory function and constructor to the OpenAIVectorStoreMixin
• Adding recording/replay support for the file_processors API in the integration test recording system, so that file processor output is captured during record mode and replayed
  deterministically in CI without running the actual processor

Test plan

Manual verification (local, inline::docling)

Save the following config to ~/.llama/distributions/providers-run/config.yaml:

config.yaml

version: 2
apis: [file_processors, files, vector_io, inference]
providers:
  file_processors:
  - provider_id: docling
    provider_type: inline::docling
    config:
      default_chunk_size_tokens: 800
      default_chunk_overlap_tokens: 400
  files:
  - provider_id: localfs
    provider_type: inline::localfs
    config:
      storage_dir: /tmp/llama-test/files
      metadata_store:
        table_name: files_metadata
        backend: sql_default
  vector_io:
  - provider_id: faiss
    provider_type: inline::faiss
    config:
      persistence:
        namespace: vector_io::faiss
        backend: kv_default
  inference:
  - provider_id: sentence-transformers
    provider_type: inline::sentence-transformers
    config:
      trust_remote_code: true
storage:
  backends:
    kv_default:
      type: kv_sqlite
      db_path: /tmp/llama-test/kvstore.db
    sql_default:
      type: sql_sqlite
      db_path: /tmp/llama-test/sql_store.db

Start the server:

LLAMA_STACK_LOGGING=providers=debug llama stack run ~/.llama/distributions/providers-run/config.yaml --port 8321

Upload a PDF, create a vector store, attach the file, and search:

FILE_ID=$(curl -s http://localhost:8321/v1/files \
  -F purpose=assistants \
  -F file=@tests/integration/responses/fixtures/pdfs/llama_stack_and_models.pdf | jq -r '.id')

VS_ID=$(curl -s http://localhost:8321/v1/vector_stores \
  -H "Content-Type: application/json" \
  -d '{"name":"test","metadata":{"embedding_model":"sentence-transformers/nomic-ai/nomic-embed-text-v1.5","embedding_dimension":768}}' | jq -r '.id')

curl -s "http://localhost:8321/v1/vector_stores/$VS_ID/files" \
  -H "Content-Type: application/json" \
  -d "{\"file_id\":\"$FILE_ID\"}"

curl -s "http://localhost:8321/v1/vector_stores/$VS_ID/search" \
  -H "Content-Type: application/json" \
  -d '{"query":"How many experts does Llama 4 Maverick have?","max_num_results":3}' | jq '.data[].content[].text'

Verify:

• Server logs show Using FileProcessor API to process file
• Search results return markdown-formatted content (docling output) with first result containing "128 experts"

[github-actions]

Recording workflow completed

Providers: gpt, azure

Recordings have been generated and will be committed automatically by the companion workflow.

View workflow run

Fork PR: Recordings will be committed if you have "Allow edits from maintainers" enabled.

[github-actions]

✅ Recordings committed successfully

Recordings from the integration tests have been committed to this PR.

View commit workflow

[franciscojavierarceo]

nice