Separate missing evidence from uncertainty

A lot of the rankings have "unclear" as a label. This is ambigious.

“Unclear” currently combines:

- no CBD/THC data;
- genuinely conflicting evidence;
- and nine cases where both drivers are clearly positive.

These require three different labels.

Start by locating where the `unclear` ranking label is defined and rendered (search for the literal label in source, config, or seed data). Then inspect the validation/normalization path that accepts label values, since it will need to allow three distinct states. Update the relevant definitions and any downstream display text, then run the project’s test/build checks and spot-check a few rankings to confirm the old ambiguous label is no longer used.

索引モデルが issue の本文から書いたものです。