Ignore character modifying codes in the prompt.

[frm]

Why:

• If a user set GEOMETRY_SYMBOL_PROMPT="\e[1mX\e[0m", it would be expected that X was bold.
• However, even though X would be bold, zsh would assume a much larger prompt character size of 11, which would in turn, mess up with the autocomplete.

This change addresses the need by:

• Performing substitution on the prompt symbol before counting the characters.

Notes:

• This should fix the issue brought up by @n-a-t-h-i-a-s in #3. See this comment.
• I'm not completely satisfied with this solution since it doesn't fully address the issue. It's very specific to this use case. However, as more use cases come by with those types of errors, we can refine it.
• It might be worthwhile to look at what other prompts are doing, such as adam2 in this Stack Overflow question.

[frm]

One possible alternative would be to recommend users to use other alternatives for bold characters. %BX%f would be equivalent and would allow us to use a substitution mechanism similar to the adam2 one I linked. It feels a bit better do that instead of this, which feels a bit hackish.

It would a matter of changing one line to

local symbol_width="${#${(S%%)prompt_symbol//(\%([KF1]|)\{*\}|\%[Bbkf])}}"

[jedahan]

I think this refactor is generally better than our current state, both in that it helps a user, and is easier to read with the split up variables. We can revisit in the future but I am fine with this hack being merged sooner than later.

[frm]

Here's a preview of the support:

When merging I'd rather squash and keep the first commit message, as it provides context for everything, when looking through the git log 😄

[jedahan]

Ahh squash & merge makes sense. This looks good to me