fix(intelligence): unify rule-based importance scoring

[knqiufan]

Summary

• Closes [Enhancement]: Unify rule-based importance evaluation with six-dimension weighted scoring #1141.
• Refactors rule-based importance evaluation to reuse the six-dimension weighted scoring model.
• Adds weighted_total to importance breakdown and shares weighted aggregation with LLM criteria synthesis.
• Expands rule-based keyword signals for Chinese content and full-width punctuation.

Test plan

• python -m pytest tests/unit/intelligence/test_importance_evaluator.py
• python -m pytest tests/unit/intelligence
• git diff --check

[wayyoungboy]

Requesting changes because the current rule-based scoring path drops existing metadata/context importance signals.

The new rule-based path delegates to dimension scoring, but that scoring path does not appear to consume the existing metadata.priority, metadata.tags, or context.user_engagement signals. A direct probe on the current head with neutral content plus high-priority/profile/high-engagement metadata/context returns 0.0; the previous rule fallback assigned a positive score from those signals. The new tests patch the dimension scorer, so they do not exercise the real evaluator behavior.

Please carry the metadata/context boosts into the real dimension scoring path, or keep the old signal path covered by a test that calls the real evaluator instead of patching it.

[knqiufan]

@wayyoungboy Thanks for the review! Fixed in 0d0da9a.

The rule-based path now carries the existing metadata/context signals into the real dimension scoring path:

• metadata.priority and context.user_engagement contribute through relevance.
• metadata.tags contributes through novelty.
• Added a regression test that calls the real evaluate_importance path with neutral content plus high-priority/profile/high-engagement metadata/context and verifies the legacy positive score is preserved without patching the scorer.

Verified:

• python -m pytest tests/unit/intelligence/test_importance_evaluator.py
• python -m pytest tests/unit/intelligence
• git diff --check

[wayyoungboy]

Thanks for the follow-up. This is still not ready to merge while the previous CHANGES_REQUESTED review is active.

Please request re-review after confirming that the rule-based path carries the existing metadata/context importance signals into the six-dimension scoring path, with regression coverage for those signals. Once that review is cleared, this can be reconsidered for merge.