Improve compile-time if conditions handling

[GuillaumeGomez]

Fixes #115.

It finally allows to write:

{% if x is defined && x == 12 %}

without having compile-time errors because of the x == 12 part. :)

[GuillaumeGomez]

Fixed fmt. :)

[Kijewski]

Will review in the next few days, today I'm wiped. :)

[Kijewski]

Interesting, why didn't clippy like that? Because Splitter is not pub, so it's known that there won't be any unsized Splitters?

[GuillaumeGomez]

Here's the output:

warning: `?Sized` bound is ignored because of a `Sized` requirement
  --> rinja_parser/src/memchr_splitter.rs:10:20
   |
10 | impl<T: Splitter + ?Sized> Splitter for &T {
   |                    ^^^^^^
   |
note: `T` cannot be unsized because of the bound
  --> rinja_parser/src/memchr_splitter.rs:10:9
   |
10 | impl<T: Splitter + ?Sized> Splitter for &T {
   |         ^^^^^^^^
   = note: ...because `Splitter` has the bound `Copy`
   = note: ...because `Copy` has the bound `Clone`
   = note: ...because `Clone` has the bound `Sized`
   = help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#needless_maybe_sized
   = note: `#[warn(clippy::needless_maybe_sized)]` on by default
help: change the bounds that require `Sized`, or remove the `?Sized` bound
   |
10 - impl<T: Splitter + ?Sized> Splitter for &T {
10 + impl<T: Splitter> Splitter for &T {
   |

[Kijewski]

I'm not super happy that the Expr need to get cloned, but I don't see a better implementation. Adding e.g. an enum { &Expr | Box<Expr> } type would make the code unreadable, because the change would be viral: any &Expr would need to be changed down the line.

And since the conditions should not be excessively big (ensured by our nesting level check), this should be fine.

Thanks for the PR! :)

[GuillaumeGomez]

I'm not super happy about it either, that's why I added the contains_bool_lit_or_is_defined field to only clone when it's actually needed to limit performance impact.