Refactor: centralise critique module, switch to config.json, add provider/model config, concurrency and agent profiles

[justinlietz93]

Motivation

• Reorganise and modularise the critique functionality under a dedicated critique_module package and standardise on a single JSON configuration to simplify deployment and provider configuration.
• Add support for Anthropic as a first-class provider and centralise provider model lookups to avoid hard-coded defaults and surface configuration errors early.
• Enable per-agent provider overrides and controlled parallelism to improve throughput and allow multi-provider multi-agent workflows.
• Provide utilities for indexing internal PDFs into the ArXiv vector store to support offline/internal reference material ingestion.

Description

• Moved many core modules into src/critique_module and updated imports across the codebase to use the new package (examples: content_assessor, council_orchestrator, reasoning_agent, output_formatter, pipeline_input, etc.).
• Replaced the previous YAML configuration loader with a JSON-backed critique_module.config_loader and updated config.json adding anthropic config, agent_profiles, and a concurrency.max_parallel_calls option, while removing config.yaml and the old src/config_loader.py.
• Introduced providers.model_config for centralised model/provider configuration access, updated provider clients (openai, gemini, anthropic/claude) to resolve models from config.json and raise clear errors when models are not configured, and adjusted the provider call paths to use these accessors.
• Implemented concurrency controls and per-agent provider selection in the council orchestrator and thesis builder using asyncio semaphores, and added logic to build per-agent API configurations via _build_agent_config and _resolve_agent_provider in council_orchestrator.
• Added src/__main__.py module entrypoint and a new CLI utility src/arxiv/cf_loader.py for indexing internal PDF corpora into the ArXiv vector store.
• Updated numerous application and infrastructure modules to reference the relocated pipeline_input and config loader paths and removed obsolete React context files and the legacy config_loader module.

Testing

• Ran the automated unit test suite with pytest -q and all tests passed.
• Ran the code style checks with flake8 and no style errors were reported.
• Executed a lightweight smoke test of the thesis builder entry path using python -m src.syncretic_catalyst.thesis_builder with a mocked config and it completed the agent scheduling portion without runtime exceptions.

---

Codex Task