Optimize FixedArray reverse iteration

[mizchi]

Summary

Optimize reverse iteration helpers on FixedArray by using unchecked reads in the already bounds-controlled loops.

This covers:

• FixedArray::rev_each
• FixedArray::rev_eachi
• FixedArray::rev_fold
• FixedArray::rev_foldi

Benchmarks

Command: moon bench -p moonbitlang/core/builtin -f array_locals_bench_test.mbt --target native --release

Benchmark	Before	After
FixedArray::rev_each n=100000	24.39 us	4.76 us
FixedArray::rev_eachi n=100000	9.51 us	9.34 us
FixedArray::rev_fold n=100000	24.22 us	4.99 us
FixedArray::rev_foldi n=100000	49.48 us	9.75 us

Validation

• moon fmt
• moon test builtin --target all
• moon check builtin --target all --no-render
• moon info
• moon bench -p moonbitlang/core/builtin -f array_locals_bench_test.mbt --target native --release

[Copilot]

Pull request overview

Optimizes FixedArray reverse-iteration helpers in the MoonBit core standard library by switching to unchecked element reads (unsafe_get) inside already bounds-controlled loops to reduce overhead.

Changes:

• Use unsafe_get inside FixedArray::rev_each, FixedArray::rev_eachi, and FixedArray::rev_fold loops.
• Refactor FixedArray::rev_foldi to an explicit (read, index, acc) loop and use unsafe_get for the reverse read.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[bobzhang]

Three of this PR's four optimizations have already landed on main. The remaining piece — rev_foldi's per-iteration len - i - 1 — is now completed in #4117 with fresh benchmarks against current main (native 1.33x, js 1.21x, wasm-gc 1.15x), keeping the range form the sibling functions use (for i in self.length()>..0; index = 0, acc = init), which measures identical to the C-style loop proposed here. Your index-pairing regression test is adopted there with credit. If #4117 lands, this PR can be closed in its favor.