epic: dynamic model discovery with per-provider API listing and 1-day cache

[bug-ops]

Goal

Replace the static `list_models()` returning only the configured model with a dynamic
`list_models_remote()` async method that queries each provider's API and returns the
full set of available models. Results are cached on disk with a 1-day TTL so repeated
calls are cheap.

Motivation

• `/model` command and TUI model picker currently show only the single configured model.
• Users cannot discover what models are available without consulting external docs.
• Ollama users especially benefit from seeing all locally pulled models.

Scope

Provider	Endpoint	Auth
Ollama	`GET /api/tags`	none
Claude	`GET https://api.anthropic.com/v1/models\`	`x-api-key`
OpenAI	`GET https://api.openai.com/v1/models\`	Bearer
CompatibleProvider (Gemini, Grok, Groq, vLLM…)	`GET {base_url}/models`	Bearer
Candle	static (loaded model name)	—

Child Issues

• feat(llm): add async list_models_remote trait method and disk cache with 1-day TTL #992 — `list_models_remote` trait method + disk cache infrastructure
• feat(llm/ollama): implement list_models_remote via GET /api/tags #993 — Ollama: `/api/tags` implementation
• feat(llm/claude): implement list_models_remote via GET /v1/models #994 — Claude: `/v1/models` implementation
• feat(llm/openai): implement list_models_remote for OpenAI and CompatibleProvider #995 — OpenAI + CompatibleProvider: `/v1/models` implementation
• feat(llm): wire list_models_remote into AnyProvider, RouterProvider, ModelOrchestrator #996 — `AnyProvider::list_models_remote` dispatch + orchestrator/router aggregation
• feat(core/tui/cli): integrate dynamic model listing into /model command and TUI model picker #997 — `/model` command and TUI model picker integration