Implement minimal interpreter for new pipeline

The interpreter should be implemented for the new pipeline. This roughly consists of the following steps:

- [ ] Create a data type for the primitive operations in the interpreter. We should keep this limited to generate and permute, as the other operations have default implementations in terms of these. That will also allow us to test the default implementations :)
- [ ] Implement the DesugarAcc type class
- [ ] Integrate with new schedule (#467)
- [ ] Create a Kernel data type, containing a function performing the actual computation
- [ ] Implement compilation step (i.e., a function from Cluster to Kernel)