Unify Attrs and AttrsOwned

[geieredgar]

Currently, there exist the borrowing Attrs and the owning AttrsOwned, which only differ in their usage of Family vs FamilyOwned. In theory, this allows passing Attrs without having to allocate. In practice, we have the following situation:

1. FontSystem::get_font_matches wants to prevent an allocation in the most-common case that a cached match already exists, but it can't (see

   cosmic-text/src/font/system/std.rs

   Line 130 in 1bc198f

   .entry(AttrsOwned::new(attrs))

   ), because we cannot implement Borrow<Attrs> for AttrsOwned.
2. We need to store AttrsOwned in AttrsList, so it is almost impossible to not have an Attrs turn into an AttrsOwned somewhere down the line.

This PR removes the old Attrs, renames AttrsOwned to Attrs and adds an AttrsBuilder, which has the same helper methods as the old Attrs. This PR also includes several changes to prevent allocations in many cases:

• FamilyOwned uses a Cow<'static, str> instead of String, allowing the use of string literals without allocating.
• AttrsBuilder::family takes an impl Into<FamilyOwned>, which is only implemented for FamilyOwned or Family<'static>, so users need to explicity use FamilyOwned::new if they want to use a non static Family.
• Attrs implements AsRef<Attrs> and From<&Attrs>, allowing methods that might need to store the Attrs to accept both values and references of Attrs, while at the same time preventing an extra allocation when a value is passed.
• As a side effect of Attrs now not being Copy, users have to explicitly use clone, which makes potential allocations more obvious.

This PR also changes FontSystem::get_font_matches to take an impl AsRef<Attrs> + Into<Attrs> parameter, which it only clones if (a) the passed argument is a reference and (b) the cache does not contain an equivalent Attrs. So this change leads to probably less allocations overall.

[jackpot51]

This now has conflicts after merging #97

[geieredgar]

The conflicts are resolved, but the CI fails right now, caused by duplicate syn dependency, unrelated to my changes.

[Imberflur]

FWIW point 1 can probably be addressed via using hashbrown with HashMap::entry_ref + Equivalent.

[genusistimelord]

I think the changes here seem reasonable as it will help simplify the overall usage of Attr's.

My only Concern is how will this affect the overall performance since we are moving towards clones more than copy. @jackpot51 what do you think?