Implement the approved classic-state importer

[zoeyrose]

Outcome

Build a one-way, independently implemented, dry-run-first importer only for domains approved by #5; close this issue without code if the decision is a fresh start.

Scope and invariants

• Read a frozen copy/export of classic state and never modify the source.
• Validate source version, schema, record bounds, credentials/privacy, stable-ID mappings, content digest compatibility, duplicate handling, and unsupported domains.
• Map into canonical Go domain commands/transactions rather than writing replacement SQLite tables directly.
• Produce per-record results, summary counts, warnings/errors, deterministic digests, and a machine-readable approval report with secrets redacted.
• Support idempotent rehearsal, explicit operator confirmation, backup, abort thresholds, and rollback before replacement activity begins.

Acceptance criteria

• Golden fixtures are independently constructed or explicitly relicensed; GPL migration code/tests are not copied.
• Dry runs are the default and make no destination mutation.
• Repeated approved imports cannot duplicate accounts/items/currency/progress or overwrite newer destination state.
• Malformed/hostile/oversized input, mapping gaps, disk failure, and interruption are fuzzed/fault-injected.
• A production-like rehearsal meets Decide classic account, character, and world-state migration policy #5's acceptance/rollback thresholds before use.

Dependencies and parallelization

Strictly gated by #5, canonical schemas, and #39/#40. It is not allowed to influence the replacement domain model.

Licensing and migration rule

Replacement Go code is MIT and independently implemented. Preserve existing gameplay/content design choices and separately licensed data; do not translate GPL runtime source or tests unless an exact contribution has a documented rights-holder MIT grant.