Phase 9 implementation — adventure-bundled monsters

[mmacy]

Implements the phase 9 plan: an Adventure document carries its own MonsterTemplates, and the engine resolves them everywhere it resolves template ids. The milestone holds: a bundled custom template is validated, spawned through every engine path (keyed, command, wandering), fought to XP and treasure, persisted, and replayed — with the consumer call validate_adventure(adventure, load_monsters(), load_equipment()) unchanged.

Notable decisions (all pinned in the plan)

• The effective catalog builds once in GameSession.__init__, not lazily: load_game restores without re-running validate_adventure, and a doctored save with a colliding bundled id fails at restore with a typed ContentValidationError — never a first-spawn ValueError. An empty bundle returns the @cached base catalog object itself (identity asserted in tests): no copy, no behavior change for adventures that bundle nothing.
• Four one-line redirects, all draw-neutral: GameSession.spawn, the SpawnMonsters existence check, keyed-encounter resolution, and the listen check now read session.effective_monsters. The wandering spawn path needed no edit — its rows funnel through session.spawn. The 46 golden tests pass with zero golden-file diffs.
• Validation is total on the failure path: _effective_monsters returns a first-occurrence-wins dedup union plus every skipped colliding id, so a collision reports one line per id through the standard gate with no spurious "unknown monster" echo per keyed reference.
• Inline wandering-table monster_ids are now validated against the union — closing a pre-existing play-time ValueError surface (changelog Changed bullet).
• No SCHEMA_VERSION bump, no golden regeneration, no __all__ change; test_public_surface.py passes unedited. Collisions are errors, never overrides; there is no runtime registration surface — the phase 6 boundary stands.

Docs and spec impacts

Spec: the adventure model's bundled-templates clause, the frozen-data note (bundled templates are adventure data, not SRD data), roadmap entry 9. The authoring guide retitles to "Authoring custom classes, spells, and monsters" and gains a bundling section whose example is executed by the docs-example harness. The four-docstring id-domain sweep (GameSession.spawn, KeyedMonster, MonsterEncounterEntry, validate_adventure) names the effective domain. docs/adaptations.md gains no entries — this phase pins engineering seams, not rules readings.

Review provenance

One rubber-duck round by a skeptical senior reviewer with the full reading list (spec, plan, AGENTS.md, the complete diff, the touched code): verdict SOLID, a single non-blocking finding — the default-field test asserted only the dump side of the round-trip — fixed in fc115b7 by validating the dumped document back into Adventure; fix re-verified by the same reviewer, verdict confirmed. No blocking findings in either round.

Full gate on the branch: pytest 1535 passed / 58 skipped, ruff format --check and ruff check clean, pyright 0 errors, mkdocs build --strict clean.

Release 1.2.0 (plan work item 6) follows after merge per the phase 8 checklist.

https://claude.ai/code/session_018e1XMwtEpnzA4zuuPZhzaS