faster invalid utf8 escaping

[ruslandoga]

Consider replacing

ch/lib/ch/row_binary.ex

Lines 619 to 658 in 972f8d5

	@doc false
	def to_utf8(str) do
	utf8 = to_utf8(str, 0, 0, str, [])
	IO.iodata_to_binary(utf8)
	end
	@dialyzer {:no_improper_lists, to_utf8: 5, to_utf8_escape: 5}
	defp to_utf8(<<valid::utf8, rest::bytes>>, from, len, original, acc) do
	to_utf8(rest, from, len + utf8_size(valid), original, acc)
	end
	defp to_utf8(<<_invalid, rest::bytes>>, from, len, original, acc) do
	acc = [acc | binary_part(original, from, len)]
	to_utf8_escape(rest, from + len, 1, original, acc)
	end
	defp to_utf8(<<>>, from, len, original, acc) do
	[acc | binary_part(original, from, len)]
	end
	defp to_utf8_escape(<<valid::utf8, rest::bytes>>, from, len, original, acc) do
	acc = [acc | "�"]
	to_utf8(rest, from + len, utf8_size(valid), original, acc)
	end
	defp to_utf8_escape(<<_invalid, rest::bytes>>, from, len, original, acc) do
	to_utf8_escape(rest, from, len + 1, original, acc)
	end
	defp to_utf8_escape(<<>>, _from, _len, _original, acc) do
	[acc | "�"]
	end
	# UTF-8 encodes code points in one to four bytes
	@compile inline: [utf8_size: 1]
	defp utf8_size(codepoint) when codepoint <= 0x7F, do: 1
	defp utf8_size(codepoint) when codepoint <= 0x7FF, do: 2
	defp utf8_size(codepoint) when codepoint <= 0xFFFF, do: 3
	defp utf8_size(codepoint) when codepoint <= 0x10FFFF, do: 4

with one of these approaches:

• More benchmarks and better naive code Moosieus/UniRecover#1
• https://github.com/elixir-unicode/unicode/blob/main/lib/unicode/validation/utf8.ex