Add ad-hoc agent/MCP query demo and deployable MCP server

[felihong]

Summary

• demos/agent_mcp_query: answers ad-hoc questions against an Impulse silver layer by building and computing a report on the fly (channels/events/bins/virtual signals as call-time parameters, no report defined in code ahead of time and no persisted Gold table), then exposes that logic as MCP tools (list_channels, list_containers, preview_histogram, preview_histogram_2d, preview_stats, preview_point_values) with an in-notebook round trip against real data.
• demos/agent_mcp_app: a deployable Databricks App hosting the same tools as a custom MCP server, reachable from AI Playground or any MCP-compatible agent. See its README.md for setup, dependency-pin rationale, and known limitations (no per-user auth yet; distance/custom-weighted histograms gated off after finding they produce numerically incorrect results — the notebook's Section 3 has the repro).

Test plan

• Run demos/agent_mcp_query top to bottom on a serverless cluster with a fresh catalog/schema; confirm the in-notebook MCP round trip (Section 5) returns real data for all 6 tools.
• Deploy demos/agent_mcp_app per its README; confirm it's discoverable in AI Playground and that list_channels/preview_histogram/preview_histogram_2d return correct results against the deployed app.
• Spot-check the AI Playground test questions in Section 7 of the notebook.

This pull request and its description were written by Isaac.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}