normalize_newlines & text_replaced should not use a dictionary

Dictionaries are unordered

Dictionaries are unordered. If I have text that is:

    \r\n\r\r\n

And the \r -> \n replacement runs first, I get

    \n\n\n\n\n

When \r\n replacement runs, nothing changes (no \r).

Now if the operation runs in reverse (\r\n -> \n first):

    \r\n\r\r\n

becomes

    \n\r\n

And then I suspect

    \n\n

Which is different. Isn't that a pretty evil bug? Why are we using
dictionaries?

Ah, by looking at text_replaced, I am seeing we iterate over
replacements.items() sorted by length, but I don't think that's the
right way to do it.. if we have clauses that do

    \n -> \r

    \r -> \x

    \x -> \n

Then the ordering matters very much.