feat: two-phase indexation with direct thumbnail generation in indexer

Phase 1 (discovery): walkdir + filename-only metadata, zero archive I/O.
Books are visible immediately in the UI while Phase 2 runs in background.

Phase 2 (analysis): open each archive once via analyze_book() to extract
page_count and first page bytes, then generate WebP thumbnail directly in
the indexer — removing the HTTP roundtrip to the API checkup endpoint.

- Add parse_metadata_fast() (infallible, no archive I/O)
- Add analyze_book() returning (page_count, first_page_bytes) in one pass
- Add looks_like_image() magic bytes check for unrar p stdout validation
- Add lsar fallback in list_cbr_images() for UTF-16BE encoded filenames
- Add directory_mtimes table to skip unchanged dirs on incremental scans
- Add analyzer.rs: generate_thumbnail, analyze_library_books, regenerate_thumbnails
- Remove run_checkup() from API; indexer handles thumbnail jobs directly
- Remove api_base_url/api_bootstrap_token from IndexerConfig and AppState
- Add unar + poppler-utils to indexer Dockerfile
- Fix smoke.sh: wait for job completion, check thumbnail_url field

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

infra/migrations/0012_directory_mtimes.sql

@@ -0,0 +1,8 @@
+CREATE TABLE directory_mtimes (
+    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+    library_id UUID NOT NULL REFERENCES libraries(id) ON DELETE CASCADE,
+    dir_path TEXT NOT NULL,
+    mtime TIMESTAMPTZ NOT NULL,
+    UNIQUE(library_id, dir_path)
+);
+CREATE INDEX idx_directory_mtimes_library ON directory_mtimes(library_id);