[Question]: ReMe Light in 2.1.0b2: what's the timeline for the full ReMe4 roadmap (Auto-Link, tri-modal search, 4-category digest weights)?

I noticed 2.1.0b2 ships ReMe (0.4.1.4) as the memory backend (memory_manager_backend = "remelight"), which matches part of the announced ReMe4 architecture. I went through the code and compared it against the ReMe4 design described in the earlier feature-request reply. Could you share the roadmap for completing the rest?

Already in 2.1.0b2 (verified in code):

Two-layer storage: memory/ (append-only daily) + digest/ (curated)
Hybrid retrieval: vector (embedding) + BM25, RRF-fused (search_v2_step, vector_weight=0.7)
Auto-Memory + Auto-Dream (extract → integrate → topics → finish)
Knowledge graph: wikilink parsing (WikilinkHandler), graph_snapshot for the console
Progressive retrieval: expand_links, read_step with_neighbors (1-hop metadata)
Not yet implemented (from code review):

Auto-Link — no autolink step in steps/evolve/ (only auto_memory, auto_resource, compressor, dream); graph edges only come from manually written [[wikilinks]]
Tri-modal search — search is vector+BM25 dual-modal; graph is only used as result expansion, not a third ranking channel (_rrf_merge(vector, keyword) takes two lists)
4-category digest — actual buckets are personal/, procedure/, wiki/ (no knowledge/, proactive/); wiki/ is the fallback bucket for unrecognized topics
Per-type importance weights — only a uniform vector_weight: 0.7; no differentiated weights or time-decay for proactive memory
Time-range (since/until) and tag filtering in search — search takes only query/limit/min_score; frontmatter read/write exists but no tag filter param
Questions:

Is Auto-Link planned for an upcoming beta? (It's the piece that makes the graph "self-evolving".)
Will search be upgraded to true tri-modal fusion, or is graph-as-expansion the intended design?
Any ETA for knowledge//proactive/ buckets and per-type retrieval weights?
Thanks — ReMe is a great direction, just want to know what to expect. 🙂