lightrag-memgraph: vector query() silently drops dead candidates with no signal

[antejavor]

What's wrong

MemgraphVectorStorage.query() in integrations/lightrag-memgraph/src/lightrag_memgraph/vector_impl.py silently drops any vector-search candidate that no longer exists (dead due to Memgraph's deferred GC after a delete) — with no signal to the caller that fewer than top_k genuinely-live results were found.

If a bulk delete just ran, a query could return fewer than top_k (or zero) live results with no indication the shortfall is due to GC lag rather than a genuine lack of matching content. LightRAG's retrieval pipeline (and any other caller) has no way to distinguish "nothing else was similar enough" from "results were silently discarded".

Acceptance criteria

• query() logs (at least at debug/warning level) when the raw vector_search.search() hit count differs from the live-candidate count returned, including the counts
• Document this behavior (README or docstring) as a known Memgraph limitation so callers know under-return is possible

Blocked by

None - can start immediately