feat(vortex-array): add Interleave array encoding

[joseph-isaacs]

Adds an Interleave array: a lazy, random-access gather of N value arrays into one array, taking output row i from values[array_indices[i]][row_indices[i]].

It is the random-access analog of the Merge skeleton: instead of consuming each branch in cursor order under a selector, row_indices names an explicit position within the selected value, so rows may be reordered, skipped, or repeated. Merge is the special case where each value is consumed front-to-back exactly once.

Layout (mirrors Merge)

• vortex-array/src/arrays/interleave/mod.rs — the encoding (Interleave / InterleaveArray / InterleaveData), InterleaveArrayExt accessors (num_values, value, array_indices, row_indices), the Interleave::check invariant source of truth, try_new, the VTable, OperationsVTable::scalar_at (direct gather), ValidityVTable::validity (inner non-nullable interleave of the values' validities), and oracle-backed tests.
• vortex-array/src/arrays/interleave/execute/{mod.rs,bool.rs} — value-type dispatch plus the boolean gather kernel.
• vortex-array/src/arrays/mod.rs — module registration.
• vortex-array/src/builtins.rs — ArrayBuiltins::interleave constructor (+ an ExprBuiltins TODO), mirroring ArrayBuiltins::merge.

Spec / invariants

• N value children + two selector children: array_indices and row_indices, both non-nullable unsigned integers of equal length (the output length).
• array_indices[i] < values.len() and row_indices[i] < values[array_indices[i]].len() — per-row bounds depend on selector values, so they are a runtime precondition checked in the execution kernel rather than at construction.
• All values share a logical type up to nullability; the output type is that shared type with the union of the values' nullabilities.
• Interleave::check is the single source of truth for these invariants and is used by both try_new and the VTable::validate hook.

Selector

array_indices is always a non-nullable unsigned integer indexing into values (the boolean two-value special case was intentionally dropped). The boolean-value execute kernel handles the integer-selector path directly, so multi-value interleaves execute end to end; only non-boolean value types remain unimplemented (they construct but panic on execute, dispatched on value type).

Checks

Run with RUSTC_WRAPPER= (sandbox sccache note in AGENTS.md):

• cargo build -p vortex-array ✅
• cargo test -p vortex-array --lib interleave → 13 passed ✅
• cargo +nightly fmt -p vortex-array (+ --check) ✅
• cargo clippy -p vortex-array --all-targets --all-features ✅

Follow-ups

• Non-boolean value kernels (primitive/varbin/etc.).
• An interleave expression builtin (TODO left in ExprBuiltins).

---

Generated by Claude Code

[codspeed-hq]

Merging this PR will improve performance by 14.5%

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚡ 1 improved benchmark
✅ 1512 untouched benchmarks

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Simulation	encode_varbin[(1000, 2)]	164.1 µs	143.3 µs	+14.5%

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing claude/interleave-method-6BEge (65b6a4c) with develop (e06d80b)}