Add lazy array views for large stdlib arrays

[He-Pin]

Summary

This PR adds array-level lazy views for large std.map, std.mapWithIndex, and non-constant std.makeArray results.

The old representation eagerly allocated one LazyApply1/LazyApply2 per element. For large arrays this burns allocation even when callers only force a small suffix/prefix. The new representation keeps the callback and evaluation context once per array, computes elements on demand, and caches computed values.

Important JVM/JIT/GC choices:

• Keep the old flat Array[LazyApply*] representation below LAZY_VIEW_THRESHOLD = 4096, where the JVM is usually better off with the simple monomorphic layout.
• Materialize back to Array[Eval] from asLazyArray so existing full-array stdlib consumers remain stack-safe and predictable.
• std.reverse(lazyView) materializes before reversing so the original and reversed arrays share the same element thunks; this preserves trace/native callback semantics and avoids duplicate evaluation.
• Release captured evaluator/function state after full materialization.

Correctness Coverage

Added directional tests for selective forcing and a trace regression covering the reverse-cache behavior for all three lazy view subclasses:

• MakeArrayArr
• MappedArr
• MappedWithIndexArr

Also added bench/resources/sjsonnet_suite/lazy_array_sparse_indexing.jsonnet, a reproducible sparse-indexing workload for the benchmark harness.

JMH

Environment: JMH 1.37, Mill JVM zulu:21, -wi 3 -i 5 -w 1s -r 2s -f 1 -tu ms. Baseline is upstream/master at 3a9a4928; benchmark files introduced by this PR were passed to master by absolute path for the baseline run. This sweep was rerun after the lazy-array hot-path fix in 65840c5c.

Benchmark path	master	this PR	Result
bench/resources/sjsonnet_suite/lazy_array_comprehension.jsonnet	27.772 ± 9.548 ms/op	19.945 ± 2.908 ms/op	1.39x faster
bench/resources/cpp_suite/realistic2.jsonnet	44.530 ± 0.685 ms/op	39.766 ± 0.661 ms/op	1.12x faster
bench/resources/go_suite/foldl.jsonnet	0.070 ± 0.001 ms/op	0.072 ± 0.001 ms/op	neutral
bench/resources/sjsonnet_suite/lazy_array_sparse_indexing.jsonnet	20.023 ± 4.123 ms/op	3.488 ± 0.135 ms/op	5.74x faster
bench/resources/go_suite/reverse.jsonnet	6.434 ± 0.224 ms/op	5.001 ± 0.150 ms/op	1.29x faster
bench/resources/go_suite/base64_byte_array.jsonnet	0.749 ± 0.037 ms/op	0.748 ± 0.042 ms/op	neutral

Earlier GC profiler on the directional lazy-view benchmark:

Benchmark path	master alloc	this PR alloc	Result
sjsonnet/test/resources/new_test_suite/lazy_array_views.jsonnet	12,918,912.631 B/op	1,728,840.113 B/op	86.6% less allocation

Hyperfine vs jrsonnet

This is a Scala Native CLI comparison against the local jrsonnet binary at measurement time: jrsonnet 0.5.0-pre98, 80cd36a docs: abandon wide logo version. It is CLI wall-clock evidence, not JVM/JIT/GC evidence; JVM claims should be read from the JMH section above.

Command template:

hyperfine -N --warmup 20 --runs 50 \
  -n 'sjsonnet native' \
  'out/sjsonnet/native/3.3.7/nativeWorkdir.dest/native/sjsonnet.SjsonnetMain <benchmark.jsonnet>' \
  -n 'jrsonnet' \
  '/Users/hepin/IdeaProjects/sjsonnet/jrsonnet/target/release/jrsonnet <benchmark.jsonnet>'

Benchmark path	sjsonnet native	jrsonnet	Result
bench/resources/sjsonnet_suite/lazy_array_comprehension.jsonnet	73.1 ± 1.1 ms	125.1 ± 0.8 ms	1.71x faster
bench/resources/cpp_suite/realistic2.jsonnet	80.5 ± 0.9 ms	93.9 ± 0.8 ms	1.17x faster
bench/resources/go_suite/foldl.jsonnet	3.3 ± 0.1 ms	4.2 ± 0.1 ms	1.25x faster
bench/resources/sjsonnet_suite/lazy_array_sparse_indexing.jsonnet	14.0 ± 0.5 ms	24.6 ± 0.7 ms	1.76x faster
bench/resources/go_suite/reverse.jsonnet	12.5 ± 0.5 ms	18.7 ± 0.2 ms	1.49x faster
bench/resources/go_suite/base64_byte_array.jsonnet	5.8 ± 0.1 ms	14.1 ± 0.2 ms	2.42x faster

Validation

• ./mill -i 'sjsonnet.jvm[3.3.7].test'
• ./mill -i '__.checkFormat'
• ./mill -i 'sjsonnet.js[3.3.7].compile'
• ./mill -i 'sjsonnet.native[3.3.7].nativeLink'
• git diff --check
• JMH regression sweep on this PR and upstream/master: lazy_array_comprehension, realistic2, foldl, lazy_array_sparse_indexing, reverse, base64_byte_array
• Hyperfine: sjsonnet native vs local jrsonnet on the same 6 regression workloads

[He-Pin]

I think a SliceArr and RepeatedArr can be added in the later pr once this got merged to reduce the allocations

[stephenamar-db]

please rebase