Fix archived task search and SPARQL streaming

[kcsfelty]

Summary

• emit archived Task revisions as external search deletes and reject completed/already-archived replay before producer commit
• filter Diamond's asynciterator-backed RDF stream through its RDF/JS-compatible filter contract instead of casting it to a Node stream
• extend the packed process with global and Q1/P1 SPARQL, authorization, task archive history/replay/tombstone, rebuild, Qdrant, and restart assertions

Root causes

• Task canonical commits always used operation: "upsert", so an archived Task remained an ineligible stale head and its prior lexical document survived.
• Workshop cast D1QuadSource.match() to node:stream.Readable and called .pipe(). Diamond returns an asynciterator-backed RDF/JS stream, causing real triple patterns to return a safely masked 500. Empty VALUES/FILTER(false) did not touch the source and therefore passed.

Verification

• focused native SQLite producer regression: archive revision/history, replay without watermark advance, search absence, stale-head zero, rebuild/activate
• packed process: single-pass mixed lifecycle materialization, global and Q1/P1 SPARQL with owner/outsider scoping, lexical rebuild, Qdrant semantics, and restart
• canonical: 21 files, 134/134 tests, 923.04 seconds
• format, lint, typecheck, build, exact Node SQLite parity, readiness, performance, security, audit, pack, and release-boundary checks pass
• preserves 52 operations and catalog digest af455b4d6ab3560bca9c0ab48e9db7bd3cbcfa166ace875578d88d182069071f