feat: add MCP server instructions for LLM guidance

[bobby-smedley]

🎯 What does this PR do?

Adds the MCP spec's instructions field to CHIP's initialize response. When a client (Claude Desktop, Claude Code, Cursor, etc.) connects, the server returns a short "user manual" that the client typically injects into the LLM's system prompt — giving the agent cross-tool context that no single tool description could express.

How

• pkg/chip/instructions.md (new) — the content that ships to clients. Covers what Collibra is, key concepts (Assets, Business Terms, Data Contracts, Classifications), the tool categories with current names, recommended workflows, and key patterns (semantic-search-before-keyword, lineage-IDs-≠-DGC-UUIDs, don't-pre-flight-create_asset, etc.).
• pkg/chip/server.go — //go:embed instructions.md compiles the file into the binary; passed to mcp.ServerOptions{Instructions: ...} so the SDK emits it in the initialize response.
• pkg/chip/server_test.go — adds TestServer_InitializeResponseIncludesInstructions verifying the field is non-empty and references Collibra in a real client/server round-trip.

Content provenance

The instructions are derived from SKILLS.md (already in the repo), condensed to fit a token budget. Workflows and patterns trace back to existing statements in SKILLS.md and the tool descriptions — no new "best practices" were invented for this PR. Going forward, SKILLS.md and instructions.md should be kept in sync; drift will hurt agent behavior.

✅ Checklist

• My code follows the style guidelines of this project.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have made corresponding changes to the documentation (instructions.md is the new doc; SKILLS.md remains as the detailed reference).
• My commit messages follow the Conventional Commits standard.