avoid `materialize` in `ins_kernel!`

[belledon]

GenRFS/src/tree_walk.jl

Line 61 in a57be71

ecs = count.(eachcol(partition)) .+ 1

leads to a call to materialize

1518 (17 %) | 1518 (100 %) samples spent calling materialize
    ecs = count.(eachcol(partition)) .+ 1

one option would be to manually allocate ecs. we could instead flip the order in the nested for loop with elements iterated in the outer loop with ecs computed as a scalar.

there also appears to be some confusing syntax on (redundant assignment?)

GenRFS/src/tree_walk.jl

Line 69 in a57be71

k_ins[x, ej] = if ej == ei

and

GenRFS/src/tree_walk.jl

Lines 80 to 81 in a57be71

	k_ins[x, ej] = (pxej + pcj_inc + pci_dec) -
	(pxei + pcj + pci)