feat(indexer): content-based importance scoring at indexing time

[Mathews-Tom]

Summary

Add content-based importance scoring computed at indexing time, inspired by memora-lab/memory-service-public's observation-level importance scoring pattern. Each note chunk now carries a pre-computed importance_score (0.0-1.0) based on content richness signals, providing a ranking boost for well-connected, well-tagged, link-rich content without additional query-time computation.

Importance Score Formula

Four factors with equal weight (0.25 each):

Factor	Calculation	Rationale
Entity density	min(entity_count / 5, 1.0)	Notes mentioning more graph entities are more connected
Link density	min(wikilink_count / 10, 1.0)	Notes with more [[wikilinks]] are better integrated
Content length	min(word_count / 500, 1.0)	Longer notes (up to cap) tend to be more substantive
Tag count	min(tag_count / 5, 1.0)	More tags indicate broader relevance

importance_score = (entity_score + link_score + length_score + tag_score) / 4.0

Ranking Integration

The importance score provides up to 15% boost to the final composite score:

if importance > 0:
    final *= 1.0 + 0.15 * importance

This is applied as a post-multiplier after composite scoring (semantic + recency + density + activation + type weights), similar to how mode and status multipliers work.

Changes

New Files

• src/vaultmind/indexer/importance.py (34 lines) — compute_importance(content, tags, entities) function. Uses regex for wikilink extraction, word splitting for length. All factors capped at 1.0, combined as equal-weight average
• tests/test_importance.py (135 lines) — 15 tests across 4 classes

Modified Files

• src/vaultmind/vault/models.py — Added importance_score: float = 0.0 field to NoteChunk. Updated to_chroma_metadata() return type to dict[str, str | int | float] and included importance_score in output
• src/vaultmind/vault/parser.py — Computes note-level importance in chunk_note() via compute_importance() and propagates to all chunks created from that note
• src/vaultmind/indexer/ranking.py — Added importance_score: float = 0.0 to RankedResult. Extracts importance from chunk metadata in rank_results() and applies 15% boost to final score
• src/vaultmind/config.py — Added importance_scoring_enabled: bool = True to RankingConfig
• config/default.toml — Added importance_scoring_enabled = true to [ranking]

Backward Compatibility

• importance_score defaults to 0.0 on NoteChunk — existing chunks without the field rank as before
• Importance boost is multiplicative with a floor of 1.0 (0 importance = no boost)
• to_chroma_metadata() return type widened from dict[str, str | int] to dict[str, str | int | float] — strictly additive
• All existing ranking/composite tests pass unchanged

Test plan

• 15 new tests in test_importance.py across 4 classes:
  • compute_importance (10): empty/rich content, entity/link/tag/length factors, capping at 1.0, None handling, score range validation
  • NoteChunk metadata (2): importance in to_chroma_metadata(), default value
  • RankedResult field (2): default 0.0, explicit population
  • RankingConfig (1): importance_scoring_enabled default
• All existing ranking tests pass unchanged (60 in test_composite_ranking.py + test_ranking.py)
• Full suite: 981/981 tests pass, 0 regressions
• ruff check — clean
• mypy --ignore-missing-imports — clean
• Manual: re-index vault, verify importance_score appears in ChromaDB metadata
• Manual: confirm link-rich permanent notes rank higher than sparse fleeting notes