[REVERTME] Change disable_auto_batching_passes default to true (#1979)

Author: wsmoses

docs/src/tutorials/raising.md

@@ -128,11 +128,13 @@ raising).
 ```
 
 This IR has a nested loop, but that won't work nicely for GPUs/TPUs. Even for CPUs, XLA
-often doens't do a great job with loops. By default, we will attempt to raise loops to a
+often doens't do a great job with loops. We will attempt to raise loops to a
 tensor IR.
 
 ```@example raising_stablehlo
-hlo = @code_hlo compute_attractive_force(positions_ra, masses_ra, 2.0f0)
+hlo = @code_hlo compile_options=CompileOptions(;
+    disable_auto_batching_passes=false
+) compute_attractive_force(positions_ra, masses_ra, 2.0f0)
 @assert !contains(repr(hlo), "stablehlo.while") #hide
 hlo
 ```
@@ -142,7 +144,9 @@ the values are identical.
 
 ```@example raising_stablehlo
 y_jl = compute_attractive_force(positions, masses, 2.0f0)
-y_ra = @jit compute_attractive_force(positions_ra, masses_ra, 2.0f0)
+y_ra = @jit compile_options=CompileOptions(;
+    disable_auto_batching_passes=false
+) compute_attractive_force(positions_ra, masses_ra, 2.0f0)
 maximum(abs, Array(y_ra) .- y_jl)
 ```
 

src/CompileOptions.jl

@@ -212,7 +212,7 @@ function CompileOptions(;
     disable_scatter_gather_optimization_passes::Bool=false,
     disable_pad_optimization_passes::Bool=false,
     disable_licm_optimization_passes::Bool=false,
-    disable_auto_batching_passes::Bool=false,
+    disable_auto_batching_passes::Bool=true,
 )
     optimization_passes isa Bool &&
         (optimization_passes = ifelse(optimization_passes, :all, :none))