port multi combinators to the new design

[Geal]

the code is in src/multi/macros.rs. To rewrite:

• count_fixed (is it even possible?)
• length_count
• length_data
• length_value

Some functions should be rewritten to return a closure instead of applying on the data directly (code is in src/multi/mod.rs):

• count
• fold_many0
• fold_many1
• fold_many_m_n

Finally, most of those functions could be refactored, because their previous design using macros did not allow early returns from inside the loop. Since we can do that in a function, they should be easier to rewrite.

[loewenheim]

Something I ran into with length_value and length_count: The natural type for the first parser to return is usize because its return value is used as a length or count, respectively. But the current macros allow any parser in the first position that returns a number. For length_value, I’ve simulated this with a trait bound for the first parser that makes it return Into<usize>, but for length_count, that doesn’t work: in at least one place, length_count! is invoked with a first parser that returns u32, which is not Into<usize>.