feat: add pack authoring skills and runtime catalogs

[DavidJBianco]

Summary

Adds chat-first pack management and authoring support for Scenario 2.0 industry and organization packs, closes the runtime gaps discovered while designing those skills, and makes commands/eforge the sole tracked source for skill development.

What changed

• Adds canonical eforge-pack, eforge-industry-pack, and eforge-organization-pack skills plus shared authoring references.
• Extends skill installation for Claude, ChatGPT, and Codex, including bundled-reference rewrites and validation.
• Finalizes the unreleased pack schema and runtime adapters for process, application, destination, traffic, cadence, and storage catalogs.
• Strengthens pack resolution, semantic validation, provenance, bounded/no-follow loading, immutable byte snapshots, and safe copying.
• Updates the packaged finance, healthcare, technology, and Northstar Health examples.
• Removes all tracked .agents files and ignores .agents/; generated local skills are verified against fresh installer output.
• Records the completed milestone in TODO.md and the focused worklog.

Validation

• uv run ruff check .
• uv run ruff format --check .
• uv run pytest --no-cov -q: 5,696 passed, 21 skipped
• SMB slow integration test: 1 passed
• Six clean-room AI authoring trials: 6/6 passed
• Scenario 1 fixed-seed compatibility hashes unchanged
• Authored vs resolved generation output matched
• Fresh wheel and sdist contain all canonical skill and pack files, and no .agents files
• All generated skills pass skill-creator validation

Notes

• Targets dev.
• Does not bump the package version; the 2.0.0 release bump remains an integration/release step.
• Existing unrelated untracked worktree files were left untouched.