Extract provider catalog and truncation utilities into shared threadbear_services package

[Josh-writes]

Motivation

• Centralize provider metadata and endpoint-injection logic to eliminate duplicated definitions between the CLI and Flask front-ends.
• Reuse the same tool-result truncation logic across presentation surfaces so tool outputs are compacted consistently for LLM context.
• Prepare a shared service surface as a first step toward consolidating orchestration and reducing drift between cli and flask implementations.
• Add an audit and recovery plan documenting the path to a shared service-layer and CLI/Flask parity work.

Description

• Add a new threadbear_services package exposing BUILTIN_PROVIDERS, KNOWN_OPENAI_COMPAT_PROVIDERS, inject_endpoint_config, and truncate_tool_result via providers.py, text_utils.py, and __init__.py.
• Replace in-file provider lists and truncation helpers in cli/app.py and flask_chat_app.py with imports from threadbear_services and switch the code to call inject_endpoint_config(provider, merged_cfg, self.config) and truncate_tool_result(result, max_chars=budget_chars).
• Update pyproject.toml to include the new threadbear_services package in the package discovery list.
• Add docs/cli_flask_audit_plan.md containing the CLI + Flask audit and a recommended phased plan to extract shared services and achieve feature parity.

Testing

• No automated tests were executed as part of this change.
• Manual review ensured the imported symbols and function signatures match the updated call sites.

---

Codex Task