Compile time optimization

[AzurIce]

Some strategies from bevy (https://bevy.org/learn/quick-start/getting-started/setup/#enable-fast-compiles-optional):

• Use dynamic linking for ranim crate, like what bevy does
  Now this is achived by adding crate-type = ["rlib", "dylib"] for ranim package's lib, then use RUSTFLAGS="-C prefer-dynamic"(this requires no ranim_internal crate, however it affects all crates)
• Use faster linker like lld and mold
• Use nightly toolchain
• Use cranelift for codegen (but still unstable)

And at the same time, we can re-arrange our packages.

In user's code, we only rely on Scene's API, Items and Animations:

• Scene's API can be a simple trait.
• Items are simple data structures that implements Extract to Basic Item (like VItemPrimitive).
• Animations are also simple data structures that implements EvalDynamic.

So, the packages can be re-arrange into:

• ranim-core: provides Scene, Extract traits and Basic Items.
• ranim-render: provides implementation of render for Basic Items.
• ranim-cli or ranim-app or ...: use all above to provide render api, preview app, etc.
• ranim-items: ranim builtin items built on top of Basic Items
• ranim-anims: ranim builtin anims
• ...

And in user's code, we actually just need ranim-core.

But it will requires redesign and rewrite of the internal of timeline.

[AzurIce]

Using dynamic linking saves about 300ms in the linking stage (linking ranim with user's code) for ranim-examples in #81:

[AzurIce]

Closed by #94