FCE ships in-memory code graph for RAG

FCE is an interesting swing at a real bottleneck in code RAG: too much of the stack still pays database and serialization costs for every symbol query.

• –The strongest pitch is architectural, not just benchmark-driven: an in-memory SOA graph plus direct lookups is a very different tradeoff from Python wrappers around external vector or SQL stores.
• –The project looks especially relevant for local-first agent systems that need fast structural code navigation rather than pure semantic retrieval.
• –Python bindings and CLI support make it more usable than a C++-only experiment, which matters if the target audience is actually building LLM tooling around it.
• –The obvious question is robustness across non-C++ languages, since the repo itself says C and C++ handlers are the most battle-tested and other language parsers likely still have rough edges.
• –If the release holds up in real workloads, FCE could become useful plumbing for code intelligence tools that want graph-aware context retrieval without standing up another database layer.