Bad sharding comm

before optimizations:
```julia
using Reactant

@assert length(Reactant.devices()) >= 2

N = min((length(Reactant.devices()) ÷ 2) * 2, 8)

@show N

mesh = Sharding.Mesh(reshape(Reactant.devices()[1:N], 2, :), (:x, :y))
sharding = Sharding.NamedSharding(mesh, (:x, :y))

@show mesh

x = Reactant.to_rarray(reshape(collect(Int, 1:(1024*12)), 1024, 12); sharding)

function rotate(x)
y = x[1:100, :]
x[1:924, :] = x[101:1024, :]
x[925:1024, :] = y
nothing
end

@show @code_hlo shardy_passes=:to_mhlo_shardings optimize=false rotate(x)
@show @code_hlo shardy_passes=:to_mhlo_shardings rotate(x)
@show @code_xla shardy_passes=:to_mhlo_shardings rotate(x)

function pad(x)
return Reactant.Ops.pad(x, eltype(x)(0); low=[5, 0], high=[15, 0], interior=[0, 0])
end

@show @code_hlo shardy_passes=:to_mhlo_shardings optimize=false pad(x)
@show @code_hlo shardy_passes=:to_mhlo_shardings pad(x)
@show @code_xla shardy_passes=:to_mhlo_shardings pad(x)

function dus(x, y)
x[6:size(x, 1)-15, :] = y
nothing
end

y = Reactant.to_rarray(ones(Int, 1024-20, 12); sharding)

@show @code_hlo shardy_passes=:to_mhlo_shardings optimize=false dus(x, y)
@show @code_hlo shardy_passes=:to_mhlo_shardings dus(x, y)
@show @code_xla shardy_passes=:to_mhlo_shardings dus(x, y)

function dus2(x, y)
x[6:size(x, 1)-15, 2:11] = y
nothing
end

y = Reactant.to_rarray(ones(Int, 1024-20, 10); sharding)

@show @code_hlo shardy_passes=:to_mhlo_shardings optimize=false dus2(x, y)
@show @code_hlo shardy_passes=:to_mhlo_shardings dus2(x, y)
@show @code_xla shardy_passes=:to_mhlo_shardings dus2(x, y)
```