Inefficient Resharding for NonSharded Inputs

```julia
begin
#= /home/avik-pal/reactant/Reactant.jl/src/Compiler.jl:1589 =#
var"##data#431" = similar(Array{Float64, 3}, (142, 142, 1))
#= /home/avik-pal/reactant/Reactant.jl/src/Compiler.jl:1593 =#
(Reactant.XLA.to_host)(XLA.synced_buffer(usbuf_21), var"##data#431", Reactant.Sharding.ShardInfo{Reactant.Sharding.NoSharding, Nothing}(Reactant.Sharding.NoSharding(), nothing))
#= /home/avik-pal/reactant/Reactant.jl/src/Compiler.jl:1596 =#
sbuf_21 = XLA.IFRT.Array(Reactant.XLA.IFRT.Client(Ptr{Nothing} @0x00000000035100c0), var"##data#431", Reactant.XLA.IFRT.Sharding(Ptr{Nothing} @0x00000000319c0030))
end
```

We should be using the RemapPlan API for this (atleast I think we should be able to use that for this purpose)