v0.3.0

What's Changed

• chore: post release by @cigrainger in #26
• Add Livebook / kino_dux section to README by @cigrainger in #27
• Add pre-production callout to README by @cigrainger in #28
• Add Dux.exec/1 for raw DDL execution by @cigrainger in #29
• Remove Dux.Flame.start_pool/1, document FLAME setup by @cigrainger in #30
• Fix asof_join doc group typo by @cigrainger in #31
• Add cond/if/in to Query macro, document function pass-through by @cigrainger in #32
• Distributed perf hardening: 3-12x faster fan-out and merge by @cigrainger in #33
• Streaming merger: delegate to DuckDB (11.5x faster at high cardinality) by @cigrainger in #34
• Use IPC instead of Elixir rows for table replication (15.9x faster) by @cigrainger in #35
• Build graph adjacency in DuckDB instead of Elixir MapSets by @cigrainger in #36
• FLAME end-to-end: partitioner fix, execution metadata, setup callback by @cigrainger in #37
• Per-worker stats in meta, fix status for FLAME by @cigrainger in #38
• perf: eliminate data round-trip in Backend.query/2 by @cigrainger in #40
• feat: connection pool with persistent_term dispatch by @cigrainger in #41
• feat: SQL macros — Dux.define, define_table, undefine, list_macros by @cigrainger in #42
• feat: shuffle hardening — memory limits, spill-to-disk, exchange serialization, skew mitigation by @cigrainger in #43
• chore: upgrade ADBC 0.10 → 0.11 by @cigrainger in #45
• perf: view-based compute with schema derivation by @cigrainger in #46
• docs: FLAME cluster guide + fixes for distributed AVG rewrite, macro replay, n_rows by @cigrainger in #47

Full Changelog: v0.2.0...v0.3.0