feat(engine): add term name validation and configurable deny-list for concept/entity extraction

[pcfreak30]

Add post-extraction validation pipeline for concept and entity names to prevent NLP artifacts and generic stopwords from polluting the knowledge graph.

Three layers of defense:

• sanitizeEntityName(): strip trailing punctuation artifacts (?, :, ,, ;, smart quotes, periods) that compromise leaves on entity names
• isValidTermName(): structural rejection of parsing fragments (bracket prefixes, unmatched parens, ellipsis patterns, all-closed- class POS), plus English stopword filtering via ENGLISH_STOPWORDS
• validatedTerms(): centralized quality gate applied across all extraction paths (heuristic, provider, vision, merge) with optional user-configurable termDenyList

New config surface:

• analysis.termDenyList in swamvault.config.json for domain-specific stop words that the NLP layer cannot detect on its own
• Zod-validated via analysisConfigSchema in config.ts
• Wired through compileVault -> analyzeSource -> all extraction paths

Compound-name first-word denial: entries in termDenyList automatically reject multi-word terms starting with a deny-listed word (catches NLP artifacts like "Transcript If", "Transcript What's").

ANALYSIS_FORMAT_VERSION bumped 8 -> 9 to invalidate cached analyses.

---

@waydelyle I am sharing this in hopes you are interested in it as is, a starting point or can take it as a reference. I have been using this for over a month for my own needs and would like to see this upstreamed in some way.

All the effort is from AI and this is not my domain area as a whole, but it works for me for what I needed.