Add Lakeflow Connect skill

[jralfonsog]

Summary

ai-dev-kit has zero coverage of Lakeflow Connect today (one passing reference in the SDP skill to the ingestion_definition dict). Customers using LFC for Salesforce, Workday, ServiceNow, SQL Server, GA4, and Zerobus have no skill to follow.

This issue tracks creating a new databricks-lakeflow-connect skill covering the GA connectors as first-class plus generic patterns for SaaS and databases.

Why

• Largest unaddressed surface in ingestion. Multiple connectors, growing roadmap.
• Customer-facing daily. Every Salesforce/Workday/SQL Server CDC engagement has no skill to follow today.
• Pure skill, no MCP tool needed (CLI/API for everything; LLM can generate code).
• Aligns with the Auto Loader to SDP narrative — managed ingestion is the SaaS sibling of file-based streaming tables.

Scope

First-class (7 GA connectors + 2 generic patterns):

• Salesforce (Sales/Service)
• Workday Reports
• ServiceNow
• SQL Server (cloud + on-prem)
• Google Analytics 4 (via BigQuery)
• Zerobus
• Generic SaaS pattern (UC Connection + OAuth)
• Generic database pattern (ingestion gateway + UC Volume staging + serverless pipeline)

Pointer-only: Beta/PuPr connectors as "see roadmap" with one-line examples — they shift monthly.

Out of scope (initial PR): ad-platform connectors (Meta/TikTok/Google Ads) — Beta-only, narrow audience.

Must include: decision tree of Auto Loader vs Lakehouse Federation vs Delta Sharing vs Lakeflow Connect (most-asked FAQ internally). Link to cost estimator.

Estimated size: 600-800 lines across 4-6 files.

Approach

• New skill at databricks-skills/databricks-lakeflow-connect/
• Follow existing skill conventions (SKILL.md + reference files)
• No MCP tools — CLI/API + LLM-generated code is sufficient
• Will post in #ai-dev-kit-team Slack for alignment before opening the PR