Use `Iterators.peel`, reduce allocations, and add missing `kernelmatrix_diag!` definitions + tests

[devmotion]

This PR started with a small improvement (replace first + Iterators.drop with Iterators.peel) but lead to some more changes in the end :smile: So taken together, this PR

• replaces first + Iterators.drop with Iterators.peel
• reduces allocations in kernelmatrix! and kernelmatrix_diag! functions of KernelTensorProduct for > 2 kernels (allocations are the same for 2 kernels and are increased for 1 kernel, but in the latter case one should not construct a KernelTensorProduct in the first place)
• adds a missing definition of the binary kernelmatrix_diag! for KernelTensorProduct and a test for it to test_interface
• adds a missing definition of the binary kernelmatrix_diag! for ScaledKernel (a quick check revealed it was missing, maybe more kernels have to be fixed for tests to pass)

[theogf]

I did not know about peel it's very nice! LGTM if all tests pass