Add pruning aggregate functions

[gatesn]

Adds the pruning aggregate functions used by stats rewrites, bounded string/binary min-max variants, and the aggregate compatibility relation for matching stored bounds to requested bounds.

[codspeed-hq]

Merging this PR will degrade performance by 13.9%

⚠️ 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.

❌ 2 regressed benchmarks
✅ 1235 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	new_alp_prim_test_between[f32, 16384]	103.9 µs	118.2 µs	-12.04%
❌	Simulation	new_alp_prim_test_between[f32, 32768]	153.3 µs	181.9 µs	-15.72%

Tip

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

---

_{Comparing ngates/stats-7707/min-max-aggregate-fns (d99cb3a) with develop (ba5064a)}

[gatesn]

Review comments carried over from auto-closed #7934

#7934 was closed/merged into the stack branch while the stack was being repaired. Copying the review notes here so they remain visible on the replacement PR.

Review summaries

• robert3005 requested changes: "I think it's mostly nits but I also feel like this could use self review"
• joseph-isaacs commented
• joseph-isaacs commented
• gatesn commented

Inline review comments

• joseph-isaacs on vortex-array/src/aggregate_fn/fns/all_nan/mod.rs: Nullable, only all-valid for floats
• robert3005 on vortex-array/src/scalar_fn/fns/stat.rs: why do this if you only check the ptype afterwards?
• robert3005 on vortex-array/src/scalar_fn/fns/stat.rs: there's a array.statistics().get_as that's shorter and does the same thing?
• robert3005 on vortex-array/src/stats/expr.rs: no new usages of legacy session allowed, add a static to the test
• robert3005 on vortex-array/src/stats/expr.rs: those super look ugly but I guess they're not explicitly forbidden by lint rules
• robert3005 on vortex-array/src/stats/expr.rs: this is inconsistent, sometimes you use the shorthand and sometimes you don't?
• robert3005 on vortex-array/src/expr/expression.rs: I think you now have semantic conflict with develop
• robert3005 on vortex-array/src/aggregate_fn/fns/bounded_max/mod.rs: I think I am missing something, this is Max in the context of pruning. Why do we need another expression? Is this because we don't have aggregate -> stats expression conversion logic we talked about?
• robert3005 on vortex-array/src/aggregate_fn/fns/bounded_min/mod.rs: I think you should change this to be an enum of enum Value { min(Scalar), null(dtype) }
• robert3005 on vortex-array/src/aggregate_fn/fns/all_non_null/mod.rs: are as bool casts forbidden? This looks weird
• robert3005 on vortex-array/src/aggregate_fn/fns/all_non_nan/mod.rs: note to self: we need some shorthand for this but that's the current way to do this
• joseph-isaacs on vortex-array/src/aggregate_fn/fns/bounded_max/mod.rs: I would call this truncated max, but this does feel like a stat/pruning concept not an aggregation one?
• gatesn reply on that thread: But the whole epic is saying Stat == AggregateFn...

Bot review findings worth preserving

From Claude first pass:

• Add/confirm tests and docs for empty float AllNan / AllNonNan vacuous semantics.
• Investigate whether the Min/Max min_max() double-compute is related to CodSpeed regressions.
• The StatsRewriteRule / session #[allow(dead_code)] plumbing is probably foundation work, but should be documented or exercised when active.

From Claude second pass:

• BoundedMax may conflate Empty and Unknown in cross-shard merging; keep or add tests covering combine_partials behavior.
• try_accumulate always short-circuits for the boolean pruning aggregates; document the relationship with accumulate to avoid drift.
• Non-float AllNan / AllNonNan guards in stat_array were called out as potentially dead or inconsistent depending on intended support.
• Add comments to bounded min/max noting they delegate to min_max() and compute the unused bound for now.