add status indicators for LLM inference and provider warmup

[bug-ops]

Part of #694

Operations

ID	Location	Operation	Est. Latency
SPIN-001	zeph-llm/src/ollama.rs generate/chat	LLM inference (Ollama)	1-30s
SPIN-002	zeph-llm/src/claude.rs generate/chat	LLM inference (Claude API)	1-15s
SPIN-003	zeph-llm/src/openai.rs generate/chat	LLM inference (OpenAI API)	1-15s
SPIN-016	zeph-llm/src/orchestrator.rs warmup	Provider warmup/health check	0.5-5s

Priority: P0

These are the most visible operations — every user message triggers inference.

Approach

Status must be sent from the agent loop caller (not inside LlmProvider) since providers don't have channel access. Add send_status("thinking...") before provider.generate() / provider.chat() calls in the agent loop, and clear after response arrives.

For provider warmup, add status in the initialization sequence.