feat(ansi): precompute ASCII RGB strings for perf improvements in style.go:*ColorString()

[gregriff]

Hi! I have a bubbletea TUI for LLM chat that spends ~7% of its CPU time in ansi.foregroundColorString while rendering the streamed Markdown responses. Since this may be an important use-case (crush would benefit?), I figured it's worth discussing.

I've made this fork (code+benchmarks) and added multiple alternative implementations to the slowest part of ansi.foregroundColorString. Here is the benchmark of the original vs. alternate implementations:

I realize that most of these are ugly/unidiomatic, but using the fastest implementation brings ansi.foregroundColorString to ~2% CPU time for my TUI.

I can only seriously suggest the LUT (lookup table) implementation. This creates a global [256]string{} array and stores ASCII strings for integers 0-255, computed in init() at startup. It would remove the need for FormatUint(), integer casting, and 3 heap allocations in the *ColorString() funcs. It's much more readable than the other, albeit slightly faster implementations, with very few changes to the file. It's downsides are that it uses ~5kb to store ASCII representations of 0-255 for the life of the program, and there's a negligible startup cost to precompute them. I believe that since style.go:shift() takes care of the 0-255 bounds that this has no impact on behavior.

Let me know if this is something you'd consider and I'd be happy to make a PR.
Cheers