Implement the achievement engine and a first-slice catalog

[zoeyrose]

Outcome

Deliver the reusable native Go engine required by the full achievement epic #31, then prove it with a small representative catalog.

Scope and invariants

• Define stable achievement/category/criterion IDs, hidden/disclosure policy, non-failable invariant, progress aggregation, completion timestamps, rewards if approved, and content-version behavior.
• Subscribe to durable metrics/events through indexed registrations; never poll all achievements or embed one-off feature callbacks.
• Support checked simple criteria and approved typed CEL compound predicates; effects remain Go transactions.
• Persist per-character progress/completion idempotently and expose a structured bounded catalog/state model to Rust.
• Choose first-slice examples covering counter, distinct set, threshold, and compound behavior without settling the full M5 catalog.

Acceptance criteria

• Tests cover duplicate/reordered events, reconnect, hidden criteria, catalog revision, already-completed state, metric reconciliation, reward failure, and bounded subscriptions.
• Achievements never transition from complete to failed and completion side effects are idempotent.
• The Rust client renders first-slice state without parsing server text.
• The M5 epic Implement data-driven per-character achievements #31 links every complete catalog/content task to this engine rather than creating alternate implementations.
• No Python or Starlark is required.

Dependencies and parallelization

Depends on #36, #39, #41, #42, and #45. Full achievement catalog migration remains in #31/M5.

Independent implementation

Implement from the preserved issue-level/product behavior and independently authored fixtures. Do not copy or mechanically translate GPL server/Python code or tests. Replacement code is MIT; consumed content and assets keep their own licenses.