Optimize DisplayWidth processing path

[ainame]

Summary

• reduce overhead in processed string measurement by scanning unicodeScalars in chunks
• avoid allocating an array for multi-scalar grapheme width calculation
• keep public behavior unchanged

Benchmark

Measured locally with a temporary swiftc -O harness compiled against Sources/DisplayWidth/*.swift on the same machine before and after this change.

Case	Before	After	Change
plain-string	106776.7 ns/op	105549.7 ns/op	-1.1%
processed-ascii	136009.5 ns/op	117706.1 ns/op	-13.5%
processed-ansi-tab	34249.2 ns/op	25884.3 ns/op	-24.4%
processed-emoji	90418.0 ns/op	64631.2 ns/op	-28.5%

Testing

• swift test