ts_hexdump: move hexdump facilities to combinators rather than hand-rolled iterators

Follow-on from tailscale/tailscale-rs-old#23.

It's probably smarter/more efficient to use chained combinators to perform hexdumps instead of the current hand-rolled HexdumpIter iterator; something like:

pub type HexdumpLine = heapless::String<MAX_HEXDUMP_LINE_CHARS>;

pub fn hexdump<'a>(bytes: impl Iterator<Item = &'a u8>) -> impl Iterator<Item = HexdumpLine> {
    bytes
        .chunks(MAX_HEXDUMP_LINE_CHARS)
        .map(|line_bs| {
            let mut line = HexdumpLine::new();
            let (start, end) = line_bs.split_at(8);
            
            for b in start {
                write!(line, "{b:x} ").unwrap();
            }

            write!(line, "  ").unwrap();

            for b in end {
                write!(line, "{b:x} ").unwrap();
            }
            
            // write ascii table
            for b in line_bs {
                 line.push(get_ascii_char_for_byte(*b)).unwrap();
            }
            line.push('\n');

            line
        })
}

@dylan-tailscale and @npry need to have a discussion about this just to make sure we're aligned/understand any tradeoffs, and then implement once Dylan has his head around it.

Start by locating the current hand-rolled HexdumpIter implementation and inspect how the hexdump facilities produce their output. Review the proposed chained-combinator shape, then resolve the tradeoffs with the named maintainers before implementing it. Done means the hexdump facilities use combinators rather than the current iterator.