add method to replace that takes a function RegexMatch -> String

[epithet]

This adds the possibility to perform regex string replacements with an arbitrary function of the matching string and the capture groups as proposed in #36293:

replace("ax ay bx by", r"([ab])([xy])" =>
    RegexReplacer(m -> uppercase(m[1]) * m[2]))
# Ax Ay Bx By

I chose the name RegexReplacer for consistency with the related types Regex and RegexMatch and because I find it easy to remember that replace takes a "replacer". See also Rust's trait regex::Replacer. But this is of course up for debate and other names have been proposed.

I proposed a different API in #24598 which would solve both issues at the same time:

replace("ax ay bx by", r"([ab])([xy])") do m
    uppercase(m[1]) * m[2]
end

But this hasn't gotten any feedback yet. I tried to give some pros and cons for both variants in #36293 and I'm happy with either/both solutions. Update: in light of #40484, I don't think the do-block variant makes sense anymore as an alternative.

TODO:

• Documentation

Fixes #36293

[epithet]

Still good to have the simpler method for simpler cases:

julia> @time map(s -> replace(s, r"world" => uppercase),
                 repeat(["hello, world"], 10_000_000))
  5.487882 seconds (70.05 M allocations: 3.430 GiB, 25.61% gc time, 0.44% compilation time)
10000000-element Vector{String}:
 "hello, WORLD"
 ⋮

julia> @time map(s -> replace(s, r"world" => RegexReplacer(m->uppercase(m.match))),
                 repeat(["hello, world"], 10_000_000))
  8.028345 seconds (100.05 M allocations: 5.517 GiB, 29.23% gc time, 0.35% compilation time)
10000000-element Vector{String}:
 "hello, WORLD"
 ⋮

[epithet]

#40484 makes the do-block variant even more inconsistent and less useful. So it's at best a convenience add-on, but doesn't make sense as an alternative to me anymore.