feat: auto-detect provider when switching models via /model

[teknium1]

Summary

Fixes the confusing /model experience where typing /model deepseek-chat while on a different provider would silently keep the wrong provider, causing API errors.

Salvages the concept from PR #1177 by @virtaava, with credential awareness and OpenRouter slug mapping added.

What changed

Auto-detection in /model (cli.py + gateway/run.py):

• When no explicit provider:model syntax is given, detect_provider_for_model() finds the right provider
• Priority: direct provider with creds → OpenRouter slug match → direct provider without creds
• Bare model names get remapped to proper OpenRouter slugs (gpt-5.4 → openai/gpt-5.4)

DeepSeek as first-class provider (auth.py + config.py + models.py):

• Registered in PROVIDER_REGISTRY as an API-key provider
• DEEPSEEK_API_KEY env var, base URL https://api.deepseek.com/v1
• Static catalog: deepseek-chat, deepseek-reasoner
• Works through the existing generic API-key credential resolution path — no changes to runtime_provider.py needed

New functions in models.py:

• detect_provider_for_model(model, current_provider) → (provider, model) or None
• _find_openrouter_slug(bare_name) → full OpenRouter model ID

Examples

# Before: silently stays on openai-codex, API error
/model deepseek-chat

# After: auto-switches to deepseek provider (if DEEPSEEK_API_KEY set)
# or remaps to deepseek/deepseek-chat on OpenRouter
/model deepseek-chat

# Bare names get proper slugs
/model gpt-5.4        → openai/gpt-5.4
/model claude-opus-4.6 → anthropic/claude-opus-4.6

# Explicit syntax still works as before
/model anthropic:claude-opus-4-6

Tests

• 11 new tests for detect_provider_for_model and _find_openrouter_slug
• Updated existing tests for new behavior
• Full suite: 4548 passed, 0 regressions

Closes #1177 (concept salvaged with improvements)