perf(messages): fold ASCII highlight segments once - #178

Summary

• Replace the private hand-rolled ASCII lowering path with segment-level internal/text.Fold now that PR perf(text): add ASCII fast path to Fold #174 centralizes the ASCII fast path.
• Fold valid ASCII segments once, then split the result back into the existing one-entry-per-rune mapping.
• Keep the per-rune text.Fold fallback for non-ASCII and invalid UTF-8 so highlight indices and behavior remain unchanged.

Behavior coverage

Behavior-level tests exercise the HighlightSearchTerms package seam, including decomposed combining-mark mapping (Cafe\u0301) to preserve the original rune-to-highlight relationship.

Observed benchmark

Environment: Apple M3 Max, darwin/arm64.

Protocol:

go test ./internal/ui/messages -run '^$' -bench '^BenchmarkHighlightSearchTerms_ASCII$' -benchmem -count=6 -benchtime=500ms

In these measurements, the median was 33.5% lower and allocations decreased from 185 to 9 per operation. These are observed results on the environment above, not universal performance claims.

Context

This is the independent message-highlighting follow-up requested after PR #174 / issue #165. It is separate from the reaction-picker follow-up in PR #177.

Refs #165