feat(ollama): comprehensive model management and server integration

[Koan-Bot]

Summary

Comprehensive Ollama integration for Kōan — from REST client to Telegram-accessible model management.

What's included (7 commits)

Infrastructure

• Native Ollama REST API client (ollama_client.py) with health checks, model listing, version detection
• Unified _api_request() layer (GET/POST/DELETE) with consistent error handling
• OllamaClaudeProvider — routes Claude CLI through Ollama proxy via ANTHROPIC_BASE_URL
• Provider alias: ollama → LocalLLMProvider (alongside existing local)
• Docker Compose service template for Ollama with GPU passthrough docs

Skill: /ollama

• /ollama — Server status: version, model list, running models, configured model readiness
• /ollama list (aliases: ls, models) — Compact model listing
• /ollama pull <model> — Download models with streaming NDJSON progress
• /ollama remove <model> (alias: rm) — Remove locally stored models
• /ollama show <model> (alias: info) — Model details (family, parameters, quantization, context)
• /ollama help — Subcommand reference
• Provider-aware: only works with local, ollama, or ollama-claude providers

Process management

• System-wide Ollama detection before spawning — adopts already-running servers
• Config-gated auto-pull for missing models on startup (both local and ollama-claude providers)

Observability

• Startup banner enriched with Ollama status (version, model count, configured model)
• /status and /ping show Ollama process health when provider needs it
• Error messages include actionable hints (e.g., "Run: ollama pull ")

Provider validation

• OllamaClaudeProvider validates model availability via check_server_and_model() (not just server health)
• Auto-pull integration: missing models can be auto-pulled during startup quota check

Test coverage

• 215 new tests across test_ollama_client.py, test_skill_ollama.py, test_startup_info.py, test_pid_manager.py, test_cli_provider.py, test_ollama_claude_provider.py
• Full suite: 5786 passed (0 failures)

Closes #319, closes #305

Test plan

• Verify /ollama shows server status when provider is local or ollama-claude
• Verify /ollama list shows available models
• Verify /ollama pull <model> downloads and reports success
• Verify /ollama remove <model> deletes and reports freed space
• Verify /ollama show <model> displays model details (family, params, quantization)
• Verify /ollama help shows subcommand reference
• Verify startup banner shows Ollama info for Ollama-based providers
• Verify system-wide Ollama detection adopts running servers
• Verify auto-pull pulls missing models when enabled in config
• Verify provider-check rejects non-Ollama providers gracefully

🤖 Generated with Claude Code

[Koan-Bot]

Updated — Complete Rewrite

Force-pushed with a fresh implementation from upstream/main:

Changes

• New /ollama skill: status, list, pull, remove, help subcommands via Telegram
• Version checking: get_version() parses semver, is_available() validates >= v0.15.0
• Privacy: OLLAMA_NO_CLOUD=1 set by default (prevents cloud offloading in v0.16.2+)
• HTTP health check: start_ollama() now verifies server readiness via HTTP, not just PID
• Docs: updated provider-local.md with ollama-launch section and current model recommendations

Tests

• 113 new tests (69 skill + 44 provider/pid_manager)
• 6273 total pass

---

🤖 Updated by Kōan