[Docs] Update README to match current Provider architecture and API

[best]

Problem

README.md is out of date with the actual implementation after #34 (Provider unification). Several sections are misleading or incorrect.

Specific Issues

1. Quick Start missing registerProvider

Current README:

const agent = createAgent({
  model: createModel("openai:gpt-4o"),
  // ...
});

Actual code requires:

registerProvider(createOpenAIProvider());
const model = createModel("openai:gpt-4o", { apiKey: "...", baseURL: "..." });

2. No documentation for baseURL / apiKey configuration

createModel() accepts a second config parameter with apiKey, baseURL, temperature, etc. This is not documented anywhere in the README. Critical for users with OpenAI-compatible APIs (e.g., MiniMax, Azure, custom endpoints).

3. Provider architecture not documented

After #34, providers are built on two adapter base classes:

• OpenAIAdapter — used by OpenAI, DeepSeek, Qwen, Ollama
• AnthropicAdapter — used by Anthropic
  This architecture enables any OpenAI-compatible or Anthropic-compatible API. README should explain this.

4. Comparison table outdated

• Vercel AI SDK 6 now has agents (ToolLoopAgent), MCP support, memory tool — table still says "No"
• Claude Agent SDK comparison may also be outdated

5. Minor: Provider table should show custom/compatible providers

The provider table only lists 6 hardcoded providers. Should mention that any OpenAI-compatible or Anthropic-compatible endpoint works via baseURL config.

Acceptance Criteria

• Quick Start code actually works when copy-pasted (includes registerProvider + apiKey)
• Provider config (apiKey, baseURL) documented with examples
• Two-adapter architecture explained (OpenAI-compatible + Anthropic-compatible)
• Comparison table updated to reflect current state of competitors
• Custom/compatible provider usage documented

Priority

P2