Cargo.toml: Set codegen-units = 1 for release (#563)

This significantly improves code generation and optimization. In C,
translation units correspond to each `*.c` file, and thus things are
deliberately grouped together. In Rust, codegen units are chosen with a
heuristic and functions are assigned to a codegen unit (passed to LLVM).
Thus, to ensure related functions are in the same codegen unit (and are
thus inlineable), we should set this to 1. This will also make
optimization much more reliable, since a far away change can't suddenly
affect an optimization elsewhere.

We probably also want to look at enabling LTO (though we'll need to
enable LTO for `dav1d` for equivalent comparisons).

Cargo.toml

@@ -49,3 +49,6 @@ bitdepth_16 = []
 # unless we enable the basic optimizations (`opt-level = 1`),
 # which don't increase compile-time that much.
 opt-level = 1
+
+[profile.release]
+codegen-units = 1