feat: add researchclaw init command and config auto-detection

[jecanore]

Problem

First-time setup requires manually copying config.researchclaw.example.yaml to config.arc.yaml, then hand-editing provider URLs and API key variable names. The CLI also hardcodes --config to config.yaml, so it never finds config.arc.yaml without an explicit flag.

Solution

• Add researchclaw init — copies the example template, prompts for LLM provider (openai, openrouter, deepseek, acp), and writes config.arc.yaml with the right defaults
• Change --config default from "config.yaml" to auto-detect: tries config.arc.yaml first, then config.yaml
• When no config exists, the error message now suggests researchclaw init

Test plan

• researchclaw init creates config.arc.yaml with correct provider settings
• Refuses overwrite without --force; --force overwrites
• Auto-detection prefers config.arc.yaml over config.yaml
• Missing config prints researchclaw init hint
• Full suite passes (1260 tests, 9 new)

[Jiaaqiliu]

Hi @jecanore, thanks for this excellent contribution!

We've reviewed and merged this PR. The researchclaw init command and config auto-detection are now part of main.

A few notes on what we adjusted during integration:

• We resolved merge conflicts with our existing cli.py and config.py (MetaClaw bridge features were added in parallel).
• The pyproject.toml optional-dependencies section had a minor duplication after merge, which we fixed.

We also noticed a couple of areas for future improvement (not blocking):

• The provider string replacement in cmd_init() matches "openai" but the example config uses "openai-compatible" — this means the replacement is currently a no-op. We'll fix this in a follow-up.
• The example config is resolved from CWD (Path.cwd() / EXAMPLE_CONFIG), so researchclaw init only works from the repo root. Bundling the template inside the package would make it work after pip install.

Overall, great work — clean code, good test coverage. Thanks for contributing!