Alt trait for array

[snylonue]

Currently the branch::alt() only deals with tuples with at most 21 elements(#1144, #994) , so I write this pr trying to change it.

Trait Alt has been implemented for &'a mut [P] where P: Parser as a result, but due to the limition of my ability and rust compiler, I failed to implement the trait for [P] and P must have the same type.
For example, the code as follows failed to be compiled:

use nom::branch::alt;
use nom::character::complete::{alpha1, digit1};
fn main() {
fn parser(input: &str) -> IResult<&str, &str> {
  alt([alpha1, digit1])(input) // using an array
};

[coveralls]

Coverage decreased (-0.2%) to 82.52% when pulling 28265cb on snylonue:alt_trait_for_array into c5090ef on Geal:master.

[Geal]

I did not implement that trait for arrays for that exact reason. Even if some parsers have the same input, output and error types, their actual type might be different due to heavy use of closures and impl trait returns

[snylonue]

I don't think it is a good idea now

[LoganDark]

I did not implement that trait for arrays for that exact reason. Even if some parsers have the same input, output and error types, their actual type might be different due to heavy use of closures and impl trait returns

What about for parsers that have no impl types and are completely identical? You are forbidding a use case just because it is uncommon. :(

I just tested implementing Alt for all arrays using const generics, and it is definitely possible (tests compile when converted to arrays, etc), but it would take me hours of effort to read the tuple macros to understand exactly what the error behaviors are, so I'm going to call it here.

Just know that the alternative, having to implement "abuse alt associativity" in macros, is also very difficult and annoying. :/