perf: accelerate printable ASCII text width

[dexhunter]

measure_text_width currently runs printable ASCII through ANSI parsing and width calculation. This adds a boolean byte reduction that returns the byte length for printable ASCII when ANSI parsing is enabled. Other inputs retain the existing parser and width behavior.

On Rust 1.97.1 / Linux x86_64, a fixed five-corpus public-API microbenchmark improved from 320.91 ns to 90.79 ns (71.7% lower); a separate winner replay measured 90.58 ns. The mix includes three ASCII corpora, Unicode status text and colored status text. The latter two cases were about 7.0% and 2.8% slower, respectively; all eight fallback/tiny regression guards passed. These are function-latency measurements.

Validation passed all seven native feature configurations, formatting, clippy, release build and docs. Four added regression tests passed on both pristine and candidate code, and the evaluator checked 3,434,040 semantic cases in each of four ANSI/Unicode feature combinations.

The external autoresearch trajectory and exact source/metric bindings are available in Weco Observe.

The measurements and Observe source snapshots correspond to research commit 13501c50490950db912aa6cdba672825939708e1. Follow-up commit b63f12304da138f3ba40c9a56de858a2a057c485 only moves the same four regression tests into src/utils.rs; the measured implementation is unchanged. The seven feature configurations, formatting, clippy, release build and docs were validated again for this revision. The benchmark was not motivated by an application profile.

[djc]

Did this actually show up as some kind of bottleneck in your application?

[dexhunter]

This came from benchmarking the public measure_text_width API; I don't have an application profile showing it as a bottleneck. The reported improvement is limited to that microbenchmark.

I tried all() first. Against the original implementation, inputs with a long ASCII prefix followed by Unicode or ANSI were about 22.1% and 20.1% slower, which failed the regression limits. The fold() version passed those same checks. The range includes printable ASCII and space; is_ascii_graphic() excludes space, while is_ascii() includes control bytes.

I've moved the four regression tests into src/utils.rs in b63f123. Their assertions and the implementation are unchanged. All seven feature configurations, formatting, clippy, release build and docs pass.