Fast bounds-check for CartesianIndex ranges (#55596)

`StepRangeLen{<:CartesianIndex}` indices have been supported since
v1.11, but bounds-checking for such indices currently falls back to
iterating over the entire range. This PR adds a quick `checkindex` for
such ranges.

The performance improvement as a consequence:
```julia
julia> D = Diagonal(1:10_000);

julia> @Btime checkbounds($D, diagind($D, IndexCartesian()));
  6.697 μs (0 allocations: 0 bytes) # nightly, O(n)
  4.044 ns (0 allocations: 0 bytes) # This PR, O(1)
```

base/multidimensional.jl

@@ -730,6 +730,8 @@ end
 end
 @inline checkindex(::Type{Bool}, inds::Tuple, I::CartesianIndex) =
     checkbounds_indices(Bool, inds, I.I)
+@inline checkindex(::Type{Bool}, inds::Tuple, i::AbstractRange{<:CartesianIndex}) =
+    isempty(i) | (checkindex(Bool, inds, first(i)) & checkindex(Bool, inds, last(i)))
 
 # Indexing into Array with mixtures of Integers and CartesianIndices is
 # extremely performance-sensitive. While the abstract fallbacks support this,

stdlib/LinearAlgebra/test/diagonal.jl

@@ -1354,4 +1354,9 @@ end
     end
 end
 
+@testset "bounds-check with CartesianIndex ranges" begin
+    D = Diagonal(1:typemax(Int))
+    @test checkbounds(Bool, D, diagind(D, IndexCartesian()))
+end
+
 end # module TestDiagonal