Add lifetime param to ParseElem

[zsol]

This PR adds an 'input lifetime parameter to ParseElem and changes changes the signature of parse_elem to take self by &'input.

This change allows ParseElem implementations that return Elements by reference, avoiding a potentially expensive clone(). Case in point, the implementation of parse_elem in ParseElem for [T] now returns &'input T.

Both of these are breaking changes.

Closes #268.

[Mingun]

It seems that Clone bound is not required anymore

[kevinmehall]

I'm not sure about the second commit -- it seems weird that [u8] would give you &u8 while str gives you char. This is actually covered up quite well by match ergonomics automatically dereferencing in the examples from the tests, but are there other cases that expose the reference type and might be confusing?

Perhaps that Clone bound should be replaced with Copy? Then if you're using some non-Copy token type where you want borrowing, recommend a newtype wrapper and a custom ParseElem impl that can return whatever you want, enabled by the new lifetime parameter.

[kevinmehall]

I cherry-picked the first commit adding the lifetime param.

Instead of the second commit, I made the possibly-controversial choice to require ParseElem::Element: Copy for all implementations, not just for [T]. This unlocks improvements like #234, as well as better representing the expectation that they are cheap to copy, because they're passed around a lot.

As far as I'm aware, LibCST is the only user of rust-peg this will break, but I assume you wanted to return &Rc<...> anyway to avoid a lot of refcount manipulation, and were waiting for the lifetime param in order to enable that.

[zsol]

assume you wanted to return &Rc<...> anyway to avoid a lot of refcount manipulation,

Correct. That was the motivation for this pr