[FEAT]: implement conversation history pruning when approaching context limits

[Nithwin]

Is your feature request related to a problem? Please describe.

Problem

Long multi-turn conversations cause the []ai.Message history to grow excessively large, wasting input tokens and triggering context size (400 Bad Request) or Tokens Per Minute (TPM) rate limits.

Describe the solution you'd like

Proposed Solution

1. Add token usage threshold checks inside internal/agent/loop.go.
2. When approaching context limits, automatically truncate or summarize the oldest conversation turns (user/assistant/tool outputs) while strictly preserving the system instructions (prompt.Build()) and recent tool context.

Describe alternatives you've considered

No response

Additional Context

No response