perf(arrow/array): copy same-dictionary indices directly

[fallintoplace]

What does this change?

• Same-dictionary concatenation used to create one temporary buffer wrapper per input chunk.
• Use concatFixedWidthBuffers directly for the dictionary indices.
• Keep the dictionary-unification path unchanged for different dictionaries.

Benchmark

Command:

go test ./arrow/array -run '^$' -bench '^BenchmarkConcatenateSameDictionary$' -benchmem -benchtime=200ms -count=3

Apple M1 Pro, Go 1.26.3. The benchmark concatenates 65,536 rows with the same four-value string dictionary, split across different chunk counts.

Chunks	Before ns/op	After ns/op	Before B/op	After B/op	Allocs before/after
1	33,736	33,262	271,049	270,962	11 / 9
8	33,791	33,809	272,353	271,650	26 / 17
64	42,134	39,250	282,785	277,153	138 / 73
1,024	198,280	157,632	462,626	371,234	2,058 / 1,033
8,192	1,202,369	928,859	1,778,211	1,057,314	16,394 / 8,201

Tests

• go test ./arrow/array -count=1
• go test ./arrow/... ./internal/...

[zeroshade]

This change safely reuses the fixed-width concatenation path for same-dictionary indices and substantially reduces allocations for large chunk counts. I verified parity across index widths, slices, offsets, nulls, empty inputs, equal-but-distinct dictionaries, and dictionary unification. Array, repeated, race, vet, and LSP checks pass; CI is green.