Add AI integration framework (Ollama, LM Studio, OpenAI, Anthropic, Universal) - #28

…niversal)

Lays the groundwork for AI-powered features (server monitoring, script/mod
troubleshooting, rule-breaker detection, admin notifications) without
building any of those features yet - this is just the plumbing:

- New AiIntegrationSettings table (one row per SystemProfile): provider,
  base URL, encrypted API key, model name, enabled flag. API key is
  encrypted at rest via the existing IRconPasswordsCryptoService, same
  scheme as PanelSettings' Steam API key
- New IAiService/AiService: sends chat-style requests to whichever
  provider is configured. OpenAI, LM Studio, and a generic "Universal"
  OpenAI-compatible endpoint all share one implementation (same
  request/response shape); Anthropic and Ollama each get their own, since
  their APIs differ (Anthropic separates the system prompt out; Ollama
  uses its own native /api/chat format instead of OpenAI's)
- New "AI Integration" page, reachable via a link from Panel Settings:
  provider dropdown, conditional base URL (required for
  Ollama/LM Studio/Universal, optional override for OpenAI/Anthropic),
  masked API key with the same configured/remove pattern as the Steam key,
  model name, enable toggle, and a Test Connection button
- Test Connection intentionally works regardless of the enable toggle -
  it's an explicit, admin-initiated diagnostic action, not an automated
  AI feature firing on its own, so requiring the feature to be "on" just
  to verify credentials would be backwards

Tested end-to-end against the real OpenAI API with a deliberately invalid
key: confirmed the request reaches api.openai.com, the Bearer auth header
and JSON body are built correctly, and OpenAI's actual error response
("Incorrect API key provided...") surfaces back to the admin. Also
verified settings persist across a reload and the base URL/API key
guidance updates correctly per provider.