scorecard: NavScorecard reader (from_dict/from_json) — Track 4 Phase 0

[transfix]

Track 4 Phase 0 (scorecard plumbing): the scorecard layer could WRITE (to_dict/to_json) but not READ. Add the inverse so a recorded corpus-fitness row loads back into Python — the plumbing a selection/loss stage reads from (nothing consumes it yet; Phase 1+ wires it into training).

Because to_dict's keys are field-for-field the C++ cvc::nav::scorecard_json keys, the same reader loads a dbg_arrival_check3 --episodes / scorecard_json row directly.

• NavScorecard.from_dict(d): exact inverse of to_dict across all Track-1 field families (base + formation + coverage + drive + material_time_share + nested mean_coverage). Tolerant of a partial producer (missing keys keep dataclass defaults); ignores an embedded "rf" sub-object (the DBG scorecard_json shape) — the base reader is RF-free, the DBG side reads the RF scorecard separately.
• NavScorecard.from_json(s): json.loads + from_dict.

Tests: full round-trip (to_dict→from_dict and to_json→from_json reproduce the row exactly, all field families populated) + partial/rf-tolerance. The existing C++↔Python parity gate already pins every aggregate field, so this only adds the reader's own coverage. 13/13 pass; ruff clean.

Next (Phase 1, SELECTION): consume a recorded scorecard's composite_score in CommTrainer + formation/belief/grip as nav_scorecard selection fields.