- Extract license.identifier (e.g. CC-BY-SA-4.0) into new
  license_short column
- Warn if a license array has more than 1 entry (none seen yet)
- Include license_short IS NULL in getExistingWikiArticles query

- New --log-level flag: debug (default info), info, silent
  debug: every API request logged (method, URL, status, duration)
  info:  normal events (batch progress, entry counts, summaries)
  silent: only warnings and fatal errors
- Replaced all log.Printf/Fatalf calls with level-gated helpers
- API request timing added to queryWikiArticle, queryWikidataBatch, downloadFile
- Retries and backoff logged in debug mode

- Extract directors from infobox 'Directed by' field/list
- Extract screenwriters from infobox 'Screenplay by' list
- Extract actors from Cast section list (first link = person name)
- Upsert into people table, link via who table (profession: actor=1, director=2, screenwriter=3)
- Track processed entries with has_people flag column
- Consumer inserts people and marks has_people=1 on success

- Remove non-blocking select/default that silently dropped entries
- Channel sized to hold all pending entries (existing + SPARQL)
- Blocking send backpressures SPARQL if consumer is slow

- fetchWikiArticlesData is standalone again (re-extracted from consumer)
- -wiki-only flag skips SPARQL pipeline, runs only wiki data fetch
- Default behavior: full pipeline (SPARQL + wiki data in parallel)

- Merge fetchWikiArticles + fetchWikiArticlesData into one pipeline
- SPARQL producer fetches batches, commits each to DB, forwards resolved articles
- Wiki data consumer runs concurrently, fetching at 2s/request
- Each SPARQL batch commits independently (no global transaction)
- Rate limits respected for both Wikidata SPARQL and wiki server
- No parallel requests to either endpoint

- wikidata.go: url.PathUnescape SPARQL titles before storing
- wikiarticle.go: PathUnescape on read, PathEscape on send
- DB holds decoded names; URLs are always freshly encoded

Add has_no_wiki_article = 0 filter so entries previously marked as
having no Wikipedia article are not re-queried on subsequent runs.

- Mark entries as has_no_wiki_article=1 when Wikidata returns no result
- Also mark entries in batches that failed with HTTP errors
- Re-run populated 2705 wiki articles, 592 marked as no wiki

- Query Wikidata SPARQL in batches of 30 for entries missing wiki_article
- Store wiki_article title in imdb table
- Respect rate limits with configurable delay and retry on 5xx/429
- Skip entries that have no Wikipedia article
- Removed unique constraint on wiki_article (multiple entries can share one)