perf(utf8): reduce lossy decode output allocation

[mizchi]

Summary

Size the non-JS decode_lossy UTF-16 scratch array to the input byte length instead of twice that length: one UTF-8 byte can produce at most one UTF-16 code unit, including replacement characters for malformed input. When decoding fills the scratch array exactly, reuse it as the result instead of allocating and copying again.

This also halves peak scratch storage.

Benchmark

Native release, 1,000,000 input bytes:

Input	Before	After
ASCII	244.18 us	97.66 us
all invalid	3.39 ms	2.87 ms
mixed 1/2/3/4-byte UTF-8	830.00 us	669.27 us

Validation

• moon test encoding/utf8 --target all
• moon check
• moon info (no interface changes)

[Copilot]

Pull request overview

This PR optimizes the non-JS UTF-8 decode_lossy implementation by reducing UTF-16 scratch buffer over-allocation and avoiding an extra allocation/copy when the scratch buffer is filled exactly, improving throughput and peak memory usage.

Changes:

• Size the non-JS decode_lossy UTF-16 scratch FixedArray[UInt16] to src_length (instead of src_length * 2).
• Reuse the scratch buffer as the final result when written == src_length to avoid allocating/copying.
• Add benchmark tests for decode_lossy and include the bench test dependency.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
encoding/utf8/moon.pkg	Adds moonbitlang/core/bench as a test-only dependency to support new benchmarks.
encoding/utf8/decode_nonjs.mbt	Reduces scratch allocation for decode_lossy and reuses scratch when fully utilized to avoid extra allocation/copy.
encoding/utf8/decode_lossy_bench_test.mbt	Adds benchmark coverage for ASCII, invalid, and mixed UTF-8 inputs at 1,000,000 bytes.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.