Presize array copy consumers

[He-Pin]

Motivation

#822 gives consumers a cheap Eval copy API, but std.flattenArrays and array-separator std.join can still pay ArrayBuilder growth/copy costs when the outer array has a modest number of large child arrays.

This PR adds a guarded two-pass pre-size path for those consumers. The goal is to remove avoidable intermediate allocation in few-large-array workloads without regressing many-small-array workloads.

Constraints:

• do not force element values
• avoid many-small-array regressions
• guard total length before allocation
• keep hot paths as straight indexed loops
• keep this PR narrowly stacked on Add array eval copy API #822

Modification

Stacked on #822.

Use Arr.copyEvalTo to presize high-volume array-copy consumers:

• std.flattenArrays
• array-separator std.join

The pre-sized path uses two linear scans only when the outer part count is modest (<= 1024). Large outer arrays fall back to the one-pass ArrayBuilder + copyEvalTo path from #822.

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.Std0150FunctionsTests sjsonnet.ValArrayViewTests
• ./mill --no-server 'sjsonnet.jvm[3.3.7].test'
• ./mill --no-server 'sjsonnet.jvm[3.3.7].checkFormat'
• ./mill --no-server 'sjsonnet.native[3.3.7].nativeLink'
• git diff --check

JMH, JVM harness, compared with #822 copy-api baseline:

Benchmark	Before	After
array_copy_views	13.002 ms/op	8.454 ms/op
realistic2	see Native data	see Native data

Scala Native hyperfine, compared with #822 copy-api baseline, using Scala Native binaries, not JVM jars:

Benchmark	Before	After
array_copy_views	11.9 ms +/- 1.2 ms	10.5 ms +/- 1.0 ms
many-small fallback	7.0 ms +/- 0.7 ms	6.6 ms +/- 0.5 ms
realistic2	82.6 ms +/- 0.8 ms	82.5 ms +/- 0.7 ms

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 (#823)	source-built jrsonnet	Result
array_copy_views	9.3 ms +/- 0.2 ms	14.3 ms +/- 0.4 ms	sjsonnet 1.53 +/- 0.06x faster
realistic2	79.9 ms +/- 2.2 ms	92.9 ms +/- 1.9 ms	sjsonnet 1.16 +/- 0.04x faster

JIT / GC review:

• The second pass copies Eval references into one preallocated Array[Eval]; it does not force element values.
• totalLen is accumulated as Long and checked before allocating the final Array[Eval].
• PresizedCopyMaxParts = 1024 avoids turning many-small arrays into an always-two-pass workload.
• The fallback path preserves Add array eval copy API #822 behavior for large outer arrays.
• The hot path is simple counted while-loops plus copyEvalTo, so it stays friendly to JIT inlining and Scala Native codegen.

Rollback boundary:

• This PR only changes fully-consumed array-copy consumers.
• It does not change string join, renderer, sort, callback invocation, or global array view semantics.
• If a workload shows a many-small regression, the threshold can be lowered or the affected consumer can use the Add array eval copy API #822 one-pass path.

References

• Builds on Add array eval copy API #822 array eval copy API.