Add ZeroCopy marker and SchemaRead support

[cpubot]

We have simple zero-copy support for byte slices (&'de [u8]), but we can provide enough compile time information to make this more robust.

This PR adds a marker trait ZeroCopy that is used to provide blanket SchemaRead implementations for:

• &'de T where T: SchemaRead<'de> + ZeroCopy
• &'de [T] where T: SchemaRead<'de> + ZeroCopy

Primitive types that implement ZeroCopy are:

• u8
• i8
• [T; N] where T: ZeroCopy

#[derive(SchemaRead)] will implement ZeroCopy on structs when its TypeMeta is TypeMeta::Static { zero_copy: true }

Vs TypeMeta

#16 introduced TypeMeta which also provides compile-time zero_copy information. So, why do we need this?

TypeMeta is geared for usage inside SchemaWrite and SchemaRead implementations. Implementations can provide optimized compile-time branches depending on a type's TypeMeta. Due to limitations in the type system, TypeMeta is unusable for implementing something like a ZeroCopy marker trait. In particular, constants in generics can't be used to constrain implementations. E.g., we cannot

struct Assert<const B: bool>;
trait IsTrue {}
impl IsTrue for Assert<true> {}

impl<'de, T> ZeroCopy for T 
where 
    T: SchemaRead<'de>,
    Assert<{ matches!(T::TYPE_META, TypeMeta::Static { zero_copy: true, .. }) }>: IsTrue,

And as such we can't do anything similar to provide impl<'de, T> SchemaRead<'de> for T where SchemaRead<'de> ...TypeMeta...

Similarly, we could not use the ZeroCopy marker trait to provide blanket implementations for built in types like Vec, Box, etc. because the compiler limits us to a single blanket implementation for each type, so we'd have to get into newtypes, which would yield a terrible API.

So, there's a bit of duplication between where we advertise zero-copy, but the surface area is small since only u8, i8 and [T;N] are marked zero-copy and it's a relatively small amount of code to support. Users generally wont have to interface with any of these mechanics, so it's just a net win from a user-facing perspective.

[cpubot]

This needs to include a padding check like #19, marked as a draft until I figure out a way to do that