feat(providers): add Requesty gateway provider

[Thibaultjaigu]

Summary

Adds Requesty as a built-in provider, set up the same way as the OpenRouter entry from #179. Requesty is an OpenAI-compatible LLM gateway (700+ models, one key) at https://router.requesty.ai/v1.

Changes

• cheetahclaws/providers.py: new requesty entry in PROVIDERS next to openrouter (type openai, REQUESTY_API_KEY, max_completion_tokens cap, curated model list for the /model Tab picker), plus the module docstring. No new request logic. Everything after requesty/ goes out as the model ID, so the existing _provider_name, lookup_model_key and context window handling apply as is.
• Model IDs can be catalog paths (requesty/openai/gpt-4o-mini) or managed policy IDs with no vendor prefix (requesty/claude-sonnet-4-6). IDs ending in @eu route through EU providers only. The suffix belongs to the model ID and is not parsed like OpenRouter's @provider.
• cheetahclaws/commands/config_cmd.py: one /model example line.
• tests/test_requesty_provider.py: entry, prefix stripping, dispatch to the Requesty base URL, key from config and env, no DeepSeek-only fields on requesty/deepseek-v4-flash, max_tokens instead of max_completion_tokens, pricing and context window lookups, prompt overlay with @eu.
• Docs: README (provider list, table, gateway note, key pages, examples), docs/guides/usage.md (new Requesty section), reference.md, recipes.md, architecture.md, and the table row plus gateway note in the seven i18n READMEs. I left docs/news.md alone since release notes look maintainer written.

How to test

export REQUESTY_API_KEY=rqsty-...     # or /config requesty_api_key=...
cheetahclaws --model requesty/openai/gpt-4o-mini

Validation

• python -m pytest tests/test_requesty_provider.py tests/test_openrouter_provider.py tests/test_model_completion.py: 45 passed.
• Full suite on Python 3.12 (macOS): 2857 passed, 25 skipped, 3 failed. The 3 failures (test_kernel_sandbox::test_rlimit_as_kills_runaway_alloc, test_skill_nested::test_nested_SKILL_uppercase, test_vscode_setup::test_local_target_is_user_settings_when_the_editor_dir_exists) fail the same way on clean main on this machine and look macOS specific.
• Live run through providers.stream() with a real key: requesty/openai/gpt-4o-mini, requesty/claude-sonnet-4-6, requesty/gpt-5.4-mini, requesty/deepseek-v4-flash and requesty/gpt-5-mini@eu all streamed a reply, and a tool call came back as get_time. A direct call returned HTTP 200 with model gpt-4o-mini-2024-07-18.
• Every hardcoded model ID was checked against GET /v1/models and GET /v1/models/managed.

Disclosure: I work at Requesty. Happy to adjust anything to match project conventions.