perf(internal/utils): add ARM64 NEON min/max for 8- and 16-bit integers

[fallintoplace]

Summary

• Add ARM64 NEON min/max kernels for int8, uint8, int16, and uint16.
• Keep scalar tails and scalar fallback when ASIMD is unavailable or noasm is used.
• Add boundary, extreme-value, and benchmark coverage.

Benchmark

Apple M1 Pro, n=65536, median of 3 runs:

Type	noasm	NEON
int8	65,642 ns/op	1,433 ns/op
uint8	66,233 ns/op	1,394 ns/op
int16	65,431 ns/op	2,807 ns/op
uint16	66,756 ns/op	2,726 ns/op

Checks

• go test ./internal/utils -count=1
• go test -race ./internal/utils -count=1
• go test -tags noasm ./internal/utils -count=1
• ARM_ENABLE_EXT=bogus go test ./internal/utils -count=1
• go vet ./internal/utils
• GOARCH=amd64 go test -c -o /dev/null ./internal/utils

[zeroshade]

Verified locally on Apple M1: native NEON tests, -race, noasm parity, and an amd64 compile check all pass. Dispatch correctly guards on ASIMD detection with scalar tails and fallback.

---

This review was drafted by an AI-assisted tool and
confirmed by an Apache Arrow Go maintainer.