fix: use LLM's context window to specify agent's token limit (#689)

examples/agent/wiki.ts

packages/core/src/agent/openai/base.ts

packages/core/src/agent/openai/worker.ts

packages/core/src/agent/react/worker.ts

packages/core/src/agent/runner/base.ts

packages/core/src/memory/ChatMemoryBuffer.ts