This function signature isn't right:

https://swiftpackageindex.com/ml-explore/mlx-swift/main/documentation/mlx/mlxfast/metalkernel(name:inputnames:outputnames:source:header:ensurerowcontiguous:atomicoutputs:template:grid:threadgroup:outputshapes:outputdtypes:initvalue:verbose:)

We should define the kernel (source) and take many of the other parameters at the call site. We will redefine the mlx-c metal kernel each time (mlx::core will handle caching the metal kernel itself).