feat(array): add scalar probes with lazy child contexts

[joseph-isaacs]

Adds a probe API for repeated scalar_at

  // One-off lookup without retained preparation.
  let value = array.probe(ProbeUsage::Once).scalar_at(10, &mut ctx)?;

  // Repeated lookups share preparation until the probe is dropped.
  let mut probe = array.probe(ProbeUsage::Repeated);
  let first = probe.scalar_at(10, &mut ctx)?;
  let second = probe.scalar_at(20, &mut ctx)?;

Adds an encoding vtable.

pub trait OperationsVTable<V: VTable> {
    type ProbeState<'a>: Default + 'a;

    fn probe_scalar<'a>(
        array: ArrayView<'a, V>,
        index: usize,
        _probe: ProbeAccess<'a, '_, Self::ProbeState<'a>>,
        ctx: &mut ExecutionCtx,
    ) -> VortexResult<Scalar> {
        array.array().execute_scalar(index, ctx)
    }

    // Existing scalar hook.
    fn scalar_at(
        array: ArrayView<'_, V>,
        index: usize,
        ctx: &mut ExecutionCtx,
    ) -> VortexResult<Scalar>;
}

Encodings without preparation use ProbeState<'a> = () and the default hook. Overrides receive ProbeAccess::Once(array) or Repeated(state). Child reads use probe.slot(slot)?.execute_scalar(index, ctx)?, preserving preparation recursively.

The encoding implementations are in #9844. For PCO, assuming both indices fall in the same decoded page, that follow-up enables:

array.execute_scalar(10, &mut ctx)?; // Decodes for this lookup.
array.execute_scalar(20, &mut ctx)?; // Decodes again.
let mut probe = array.probe(ProbeUsage::Repeated);
probe.scalar_at(10, &mut ctx)?;      // Decodes and retains the page.
probe.scalar_at(20, &mut ctx)?;      // Reads from the retained page.

[codspeed-hq]

Merging this PR will regress 3 benchmarks

⚠️ 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.

⚡ 12 improved benchmarks
❌ 3 regressed benchmarks
✅ 2182 untouched benchmarks
⏩ 218 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	arrow_checked_add_u32_neon[16384]	12.3 µs	20.3 µs	-39.69%
❌	Simulation	random_i16[0.95]	80.3 µs	99.5 µs	-19.27%
❌	WallTime	words_gather_scalar_avx2[1024]	147 ns	165 ns	-10.91%
⚡	WallTime	mul_u64_nonnull_neon	21 µs	15.2 µs	+37.95%
⚡	Simulation	random_i8[0.5]	94.7 µs	71.8 µs	+31.84%
⚡	WallTime	filtered_sink_i64_avx512[NineNullsInTen]	17.2 µs	13.9 µs	+23.67%
⚡	WallTime	filtered_sink_i64_avx2[OneNullInEight]	31.1 µs	25.4 µs	+22.38%
⚡	WallTime	filtered_sink_i64_avx2[NineNullsInTen]	16.2 µs	13.4 µs	+20.7%
⚡	WallTime	filtered_sink_i64_avx512[OneNullInEight]	31.8 µs	26.6 µs	+19.33%
⚡	WallTime	mul_i64_nonnull_neon	20.2 µs	17.1 µs	+18.22%
⚡	WallTime	multiply_shapes_neon[(16384, PerRowPerRow)]	20.2 µs	17.1 µs	+17.72%
⚡	WallTime	filtered_owned_i64_avx2[OneNullInEight]	25.8 µs	22.1 µs	+16.88%
⚡	WallTime	dict_canonicalize_gt_u8_neon[1000000]	560.2 µs	487.4 µs	+14.95%
⚡	WallTime	words_gather_scalar_avx2[65536]	9.4 µs	8.2 µs	+13.54%
⚡	Simulation	allocate_drop_arrow[0]	456.9 ns	402.7 ns	+13.45%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing ji/array-probe-api (9ca43c4) with develop (8f3dfb4)}

Footnotes

1. 218 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩