[FEATURE] Add Anthropic Advanced Tool Use beta support

[theslyprofessor]

Summary

Adds support for Anthropic's Advanced Tool Use beta features via feature flag.

Closes #4713

---

Motivation

OpenCode currently loads all MCP tool definitions upfront. For power users with multiple MCP servers, this creates high token costs (50K-100K+ tokens before any work begins).

Anthropic released beta features specifically designed to address this:

• 85% token reduction with Tool Search Tool
• 37% token reduction with Programmatic Tool Calling
• Significant accuracy improvements with Tool Use Examples

This PR adds the foundation for these capabilities.

---

Changes

Beta Header Support

Added OPENCODE_ADVANCED_TOOL_USE feature flag that enables the advanced-tool-use-2025-11-20 beta:

// packages/opencode/src/provider/provider.ts
const advancedToolUseEnabled = Env.all()["OPENCODE_ADVANCED_TOOL_USE"] === "true"

const betaFeatures = [
  "claude-code-20250219",
  "interleaved-thinking-2025-05-14",
  "fine-grained-tool-streaming-2025-05-14",
]

if (advancedToolUseEnabled) {
  betaFeatures.push("advanced-tool-use-2025-11-20")
}

Usage

export OPENCODE_ADVANCED_TOOL_USE=true
opencode

---

Breaking Changes

None. Feature flag defaults to false. Without the flag, OpenCode behaves exactly as before.

---

Testing

• ✅ Typecheck passes
• ✅ No breaking changes introduced
• ✅ Feature flag defaults to off (safe to merge)

---

Next Steps

This PR lays the foundation for:

1. Tool Search Tool - defer_loading config for MCP tools
2. Programmatic Tool Calling - Code execution sandbox
3. Tool Use Examples - input_examples in tool schemas

I can implement these in follow-up PRs if this approach is approved.

---

References

• Anthropic docs: https://www.anthropic.com/engineering/advanced-tool-use
• Issue #4713: Feature request with community support
• Related: claude-code issue #12836 (100+ upvotes)

---

Author Notes

I'm a power user running OpenCode with 9+ MCP servers. This feature would save me 50K-70K tokens per session. Happy to implement the full feature set if this foundation is accepted.

Thanks for reviewing! 🚀