Migrate All Rubric Items to Named Checkers and Remove Legacy Regex Pathbacklog_item

Problem

The named rubric checkers infrastructure is in place, but most rubric items across demo KBs (blank, boyd, wardley, kb-ideas) and the pyrite KB still use plain string format. These rely on the legacy regex matching path, which silently fails when phrasing doesn't match a hardcoded pattern (84% failure rate observed pre-named-checkers).

Scope

1. Migration tooling — Add `pyrite qa migrate-rubric --kb ` command that: - Reads all rubric items (system, type, KB-level) for a KB - For each string item, suggests a named checker binding (auto-match where possible, prompt for manual binding) - Outputs the updated kb.yaml with named format items - Reports items that remain judgment-only

2. Migrate demo KBs — Convert blank, boyd, wardley, kb-ideas kb.yaml files to use named checkers where deterministic checkers exist. Mark schema-covered items with `covered_by: schema`.

3. Migrate pyrite KB — Convert the pyrite KB's own kb.yaml rubric items to named format.

4. Migrate CORE_TYPE_METADATA — Any remaining string-format items in type metadata for person, document, event, organization types (most already done).

5. Remove legacy path — Once all shipped KBs use named format: - Remove `RUBRIC_CHECKERS` regex list and `match_rubric_item()` function - Remove `_bind_params()` helper - Remove `ALREADY_COVERED_PATTERNS` and `is_already_covered()` (replaced by `covered_by: schema`) - Simplify `_check_rubric_evaluation()` to only handle dict items + plain string judgment-only

Success Criteria