perf: precompute preview highlight state

[sorafujitani]

Summary

• build preview colors and highlight regexp once per text render
• pass precomputed render state into each message write
• keep invalid regex fallback silent and unhighlighted
• add a highlight-render benchmark so the hot path is measured directly

Fixes #101

Benchmark

Command:

go test -bench=BenchmarkRenderText -benchmem ./internal/preview -count=10

benchstat /tmp/ccsession-101-before.txt /tmp/ccsession-101-after.txt:

RenderTextLargeTranscript-12              2.757m ± 1%   2.756m ± 2%       ~ (p=0.853 n=10)
RenderTextHighlightLargeTranscript-12     2.858m ± 2%   2.856m ± 1%       ~ (p=0.739 n=10)

RenderTextLargeTranscript-12              1.927Mi ± 0%   1.927Mi ± 0%  +0.00% (p=0.000 n=10)
RenderTextHighlightLargeTranscript-12     2.012Mi ± 0%   1.955Mi ± 0%  -2.83% (p=0.000 n=10)

RenderTextLargeTranscript-12              17.26k ± 0%   17.26k ± 0%       ~ (p=1.000 n=10)
RenderTextHighlightLargeTranscript-12     18.00k ± 0%   17.50k ± 0%  -2.74% (p=0.000 n=10)

Wall time is dominated by transcript loading in this benchmark, but the highlight path now avoids per-message regexp compilation and drops allocations.

Verification

go test ./...
go vet ./...
golangci-lint run
gofmt -l $(git ls-files '*.go')
git diff --check