- Parse each tags part once per import and cache in `Imp::tags_by_part`
- Cap total copied tags into the model at `MAX_TAG_BYTES` (64 MiB), charging each pair a fixed size_of::<(String, String)>() in addition to text
- Bound the cache: once `tags_capped` is set, return early before parsing or caching any subsequent part
- Tests covering deduplicated part copies, empty-tag pair budgets, and verifying unread tag parts are not parsed or cached after the budget is reached