Extend apple_core_ml_library functionality

[jiawen]

apple_core_ml_library and its companion, swift_apple_core_ml_library are convenient wrappers for getting a model deployed. However, they are mostly a wrapper around coremlcompiler generate. There doesn't appear to be a wrapper around coremlcompiler compile that produces a .mlmodelc. compile_mlmodel is surfaced through resources_support but it doesn't seem to be used.

A Bazel-based wrapper around coremlcompiler compile that automatically passes arguments like --platform and --deployment-target xx.yy.zz would be extremely helpful.

[jiawen]

For context, for my project, I just need the .mlmodelc, which I can then load via +MLModel loadContentsOfURL and friends. While the interface generated via coremlcompiler generate can be convenient, it has some annoying restrictions because the class name has to match that of the .mlpackage (so it can't start with a number, etc), and generates method names that match those of the input tensors (which now also places restrictions on them).

[jiawen]

@yyuting discovered that if you put a .mlpackage into the data attribute of a suitable objc_library target, rules_apple's bundling automatically processes it into a .mlmodelc. This is nice and automatic but also surprising. Perhaps we should add a test?