Add array eval copy API

[He-Pin]

Motivation

#821 adds slice views, but stdlib consumers still need one narrow way to bulk-copy arrays without each call site reimplementing directBackingArray / range / view branches.

This PR adds a shared Eval copy API that preserves laziness and gives fully-consumed array paths a simple JVM/Scala Native friendly implementation.

Constraints:

• copy references, not values
• do not force thunks
• keep loops simple enough for HotSpot and Scala Native
• avoid widening the API beyond current consumers
• keep the PR stacked and reviewable on top of Add lazy slice array view #821

Modification

Stacked on #821.

Add a narrow Arr.copyEvalTo API for bulk copying Eval references without forcing Val values.

This PR routes these consumers through the API:

• std.flattenArrays
• array std.flatMap
• array-separator std.join

Specialized copy paths cover concat, repeat, slice, reversed lazy views, range, and byte arrays.

Result

Verification passed:

• ./mill --no-server 'sjsonnet.jvm[3.3.7].compile'
• ./mill --no-server 'sjsonnet.jvm[2.13.18].compile'
• ./mill --no-server 'sjsonnet.jvm[2.12.21].compile'
• ./mill --no-server 'sjsonnet.jvm[3.3.7].test.testOnly' sjsonnet.ValArrayViewTests sjsonnet.Std0150FunctionsTests
• ./mill --no-server 'sjsonnet.jvm[3.3.7].test'
• ./mill --no-server 'sjsonnet.jvm[3.3.7].checkFormat'
• ./mill --no-server bench.checkFormat
• ./mill --no-server 'sjsonnet.native[3.3.7].nativeLink'

JMH, JVM harness, compared with #821 slice baseline:

Benchmark	Before	After
array_copy_views	16.871 ms/op	13.937 ms/op
realistic2	41.292 ms/op	41.149 ms/op
lazy_array_slice_remove	1.070 ms/op	1.067 ms/op
large_string_join	0.541 ms/op	0.539 ms/op

Scala Native hyperfine, compared with #821 slice baseline, using Scala Native binaries, not JVM jars:

Benchmark	Before	After
array_copy_views	26.1 ms +/- 0.4 ms	10.9 ms +/- 0.3 ms

copy-api ran 2.39 +/- 0.08 times faster than slice-baseline.

External performance diff, against jrsonnet built from source at 80cd36a with cargo build --release -p jrsonnet (jrsonnet 0.5.0-pre98):

Benchmark	sjsonnet Scala Native (#822)	source-built jrsonnet	Result
array_copy_views	10.5 ms +/- 0.2 ms	14.2 ms +/- 0.4 ms	sjsonnet 1.35 +/- 0.05x faster

JIT / GC review:

• The API copies Eval, not Val, so laziness and callback/error retry behavior stay intact.
• copyEvalTo(Array[Eval], offset) lets hot consumers write into one preallocated array with indexed while-loops.
• Concat materialization now copies through the same API, so concat children can provide cheap copy paths without first materializing themselves.
• Consumer code no longer repeats polymorphic branches at every call site.

Rollback boundary:

• This PR is infrastructure plus three consumers.
• It does not change string join sizing, callback invocation, sort, or renderer behavior.
• If a consumer regresses, it can move back to the previous local copy path while keeping the API for other consumers.

References

• Builds on Add lazy slice array view #821 lazy slice array view.
• Follow-up stack: Presize array copy consumers #823 presizes selected copy consumers.