Harden similarity attach and entity lifecycle

[stevesolun]

Summary

• Align incremental similarity attach defaults with the runtime graph recommendation floor and allow tagless semantic bundle queries when semantic query is explicitly requested.
• Tighten similarity CI classification and graph artifact validation so invalid edge scores or weight/final drift fail locally before CI.
• Keep README/docs graph and test counts generated from current artifacts.
• Harden ctx-monitor read Host checks and live entity deletion, including unloading live entries before catalog removal.
• Mirror added/updated skills and agents into wiki install-source paths so graph/catalog nodes remain installable.

Validation

• python -m pytest -q --no-cov ... targeted similarity/entity/dashboard suite: 239 passed.
• python src\validate_graph_artifacts.py --graph-dir graph --deep: passed with 102,928 nodes / 2,913,960 edges.
• python src\update_repo_stats.py --check: passed.
• python scripts\ci_preflight.py --profile pr: passed 15/15 gates, including ruff, mypy, pip check, non-browser pytest + coverage, docs build, clean-host/package smoke, graph validation, build, and twine check.

Disk/artifact hygiene

• Pruned local Git LFS cache (git lfs prune) and removed generated ctx temp/cache artifacts.
• Workspace is clean; no reports or review artifacts staged.