feat: model + effort switcher (dynamically fetched from the CLI)

[SergKam]

What

A toolbar switcher next to the agent switch for the session model and reasoning effort, plus a `Model · effort` item in the status bar right after the agent name.

![placement: toolbar button next to agent switch; status bar shows Agent · Model · Effort]

Nothing is hardcoded

The model list — values, display names, descriptions, and each model's valid effort levels — is fetched at runtime from the Claude CLI's stdio control-protocol `initialize` response (`response.response.models`). Verified against the installed CLI:

value	displayName	supportedEffortLevels
`default`	Default (recommended)	low, medium, high, xhigh, max
`claude-fable-5[1m]`	Fable	low, medium, high, xhigh, max
`sonnet`	Sonnet	low, medium, high, max (no xhigh)
`haiku`	Haiku	(no effort)

The effort menu rebuilds itself from the selected model's `supportedEffortLevels`; Haiku shows no effort section. Effort's fixed enum isn't enumerable via the protocol, so it's read per-model rather than listed statically.

No session reset on switch

FlyCrys already spawns a fresh `claude … --resume ` process per message. So a model/effort switch needs no reset: the chosen `--model`/`--effort` apply on the next message, conversation preserved via `--resume`. (The CLI also exposes a `set_model` control request for live processes, but we don't need it given this architecture.)

Behavior

• Eager probe at workspace init (off-thread) populates the switcher; menu shows "Fetching models…" until it lands. Empty result (CLI missing / not signed in) leaves the placeholder — no crash.
• Persists per workspace (like the agent profile). On load, a saved model/effort no longer offered (renamed model, ZDR-excluded, unsupported effort) self-heals to Default and the correction is persisted.
• Switcher menu = one icon (`preferences-system-symbolic`) with a Model section + an Effort section; selected entries checked. Switching to a model that doesn't support the current effort clears it.

Scope

• `services/cli`: `AgentSpawnConfig.effort` → `--effort`; new `ModelInfo` + `probe_models()`.
• `models/workspace_config`: persist `model_override` + `effort` (backward-compatible deserialize).
• `ui/agent_panel`: switcher menu (parameterized GActions for arbitrary model values like `opus[1m]`), spawn wiring (override ▷ profile model ▷ default), `set_models` setter.
• `workspace`: status label, eager probe thread + poll, persistence callback.

Checks

fmt ✅ · clippy `-D warnings` (lib+bins) ✅ · tests ✅ (105) · `--model default` accepted by the CLI ✅