[lpukatch]

Channel-source groups repeatedly fetch the full Dispatcharr channel and EPGData catalogs before selecting their own streams. Share one successful catalog snapshot between candidate fetching and EPG resolution for each generation, and reuse the existing group-name lookup.

Group scoping, managed-output exclusions, active-guide checks, overlap filtering, and eligible-ID-only detail lookups still apply per source. A new generation clears the shared snapshot and name caches; failed catalog reads remain retryable.

Validation covers both cache-warming orders, candidate parity, scoped detail reads, empty catalogs, transient failures, and processor reuse. A synthetic comparison across 24 sources reduced channel-catalog and EPG-catalog reads from 24 each to 1 each, with identical complete candidate output; scoped detail reads remained at 24.

Architecture documentation and the bundled agent instructions describe the cache lifetime and exclusions.

Refs #1161. Bead: teamarr-23a3.

Validation:

• ruff check teamarr/ tests/ passed.
• pytest tests/ -v -n 4 passed: 4,860 tests on Python 3.11, using isolated temporary databases.
• npm run build passed.
• 40 targeted regressions passed with an initially empty default database path.
• Full-generation smoke test completed successfully with the candidate loaded only in a benchmark process.

Catalog changes made during a generation are picked up by the next generation.