Fix llm_filter to support prompts without context_columns

[anasdorbani]

Fixes an issue where llm_filter only worked when context_columns was provided. Now it supports simple prompts without context columns.

Before (only worked with context_columns):

SELECT llm_filter(
    {'model_name': 'gpt-4o'}, 
    {'prompt': 'Is this sentiment positive?', 'context_columns': [{'data': text}]}
) FROM unnest(['I love this product!']) as tbl(text);

After (now works without context_columns):

SELECT llm_filter(
    {'model_name': 'gpt-4o'}, 
    {'prompt': 'Is paris the best capital in the world?'}
);

The function now handles both cases:

• Without context_columns: Makes a single completion request and returns the boolean result
• With context_columns: Continues to work as before with batch processing

[Copilot]

Pull request overview

This PR adds comprehensive audio transcription support to the Flock extension. The main purpose is to enable LLM functions (like llm_complete, llm_filter, llm_rerank, etc.) to process audio files by transcribing them first, then using the transcriptions as context for LLM operations.

Key Changes:

• Implemented audio transcription functionality via OpenAI's Whisper API
• Added URL/file handling utilities for downloading and processing audio files
• Extended all LLM functions to support audio context columns with automatic transcription
• Added metrics tracking system for monitoring LLM function performance

Reviewed changes

Copilot reviewed 78 out of 81 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
src/model_manager/providers/adapters/openai.cpp	Added transcription request handling for OpenAI provider
src/model_manager/providers/adapters/ollama.cpp	Added error handling for unsupported transcription in Ollama
src/model_manager/providers/adapters/azure.cpp	Added transcription support for Azure provider
src/prompt_manager/prompt_manager.cpp	Added audio transcription processing in prompt rendering
src/include/flock/model_manager/providers/handlers/url_handler.hpp	New utility for handling file downloads and base64 conversion
src/metrics/*.cpp	New metrics tracking system implementation
test/unit/prompt_manager/prompt_manager_test.cpp	Added tests for audio transcription functionality
test/unit/model_manager/model_providers_test.cpp	Updated model name and added transcription tests
test/integration/src/integration/conftest.py	Added audio file path helper and secrets setup
test/integration/src/integration/tests/functions/**/*.py	Added audio transcription integration tests across all functions

Comments suppressed due to low confidence (3)

test/unit/model_manager/model_providers_test.cpp:1

• The model version 'gemma3:4b' does not exist. The Gemma 3 family was released after your knowledge cutoff (January 2025), but typical Gemma model naming uses '2b', '7b', '9b', '27b' parameter counts. The '4b' variant is not a standard Gemma model size. Consider using a valid Gemma model like 'gemma2:2b' or 'gemma2:9b'.
  test/unit/model_manager/model_manager_test.cpp:1
• The model version 'gemma3:4b' does not exist. Please use a valid Gemma model version such as 'gemma2:2b' or 'gemma2:9b'.
  test/integration/src/integration/tests/functions/scalar/test_llm_embedding.py:1
• The model 'gpt-4o-mini-transcribe' referenced in other test files does not exist. OpenAI's transcription models use 'whisper-1' as the model identifier. Please ensure transcription model references use the correct model name.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.