serverless-migration: failure reporting flow + Genie Code install docs

[rrf-DB]

Summary

Two improvements to the databricks-serverless-migration skill, plus the repo-level glue they need.

1. Auto-file GitHub issue when migration can't complete. The Failure Reporting Protocol now produces a pre-filled issues/new URL (URL-encoded title + body) and an optional gh issue create command. The agent presents both alongside the local report path and never auto-submits. The MUST-NOT-CONTAIN list is extended (codenames, Slack IDs, go/ links, internal ticket IDs, gdoc IDs) and a defensive redaction checklist runs before write.

2. Document Genie Code Agent-mode install. A new reference doc covers the three install methods (notebook installer, shell installer, app-based workspace-wide deploy), the per-user vs workspace-wide paths, and the important serverless-no-CLI caveat for this specific skill. SKILL.md now surfaces the option at the top and instructs the agent to mention it after each migration analysis.

Changes

Skill (skills/databricks-serverless-migration/)

• SKILL.md — extended Failure Reporting Protocol (simpler decision tree: offer for ANY unmigratable case, required output template with pre-filled URL + gh command); new "Where to Run This Skill" section; reference links updated
• references/failure-reporting.md — new: redaction checklist, URL-encoding recipe, worked example, gh CLI alternative
• references/install-in-databricks-genie-code.md — new: install methods, paths, CLI caveat, follow-up Slack channels

Repo root

• .github/ISSUE_TEMPLATE/migration-feedback.md — new: matches the schema the skill emits
• README.md — list databricks-serverless-migration under Available Skills
• manifest.json — adds the two new reference files; top-level updated_at refreshed

Test plan

Validated by the local migration test bench: 16/16 cases pass (10 existing + 6 new purpose-built unmigratable workloads exercising hbase / custom-jar / py4j / sparkr / cassandra / unknown-pattern paths). Failure-report score 1.0 on every unmigratable case; zero PII leakage in the offered issue bodies.

• 6 new test workloads: synthetic-hbase-connector, synthetic-custom-jar-datasource, synthetic-py4j-jvm-reflection, synthetic-r-sparkr, synthetic-cassandra-connector, synthetic-unknown-future-api. Each expects should_trigger: true with a specific expected_error_category.
• Evaluator extended to score: referenced_report_path, offered_prefilled_url, offered_gh_cli, expected_error_category_mentioned, and a hard-fail check for PII leakage in the offered body (scoped to the offered text, not analysis chatter).
• 10/10 pre-existing cases still pass after the SKILL.md changes — zero regressions on baseline detection.

Reviewer checklist

• python3 scripts/skills.py validate passes cleanly
• Issue template renders correctly when opened via the pre-filled URL (the template=migration-feedback.md parameter resolves once this PR merges)
• Failure-reporting flow reads sensibly end-to-end

Notes

• This is a docs + protocol change. No detection-logic regressions expected.
• The skill's compatibility: Requires databricks CLI declaration is unchanged. The new Genie Code install doc explicitly flags that this is a known limitation when running inside a serverless Databricks workspace and offers three workarounds.

This pull request and its description were written by Isaac.