inlining: remove bonus for Tuple return with heap references

Added in #22210 (and earlier begun in #20853), this is no longer
necessary to avoid heap allocations, and thus serves little purpose now.

base/compiler/optimize.jl

@@ -249,7 +249,7 @@ function optimize(opt::OptimizationState, params::OptimizationParams, @nospecial
             # obey @inline declaration if a dispatch barrier would not help
         else
             bonus = 0
-            if result ⊑ Tuple && !isbitstype(widenconst(result))
+            if result ⊑ Tuple && !isconcretetype(widenconst(result))
                 bonus = params.inline_tupleret_bonus
             end
             if opt.src.inlineable

base/compiler/types.jl

@@ -44,7 +44,7 @@ struct OptimizationParams
     inlining::Bool              # whether inlining is enabled
     inline_cost_threshold::Int  # number of CPU cycles beyond which it's not worth inlining
     inline_nonleaf_penalty::Int # penalty for dynamic dispatch
-    inline_tupleret_bonus::Int  # extra willingness for non-isbits tuple return types
+    inline_tupleret_bonus::Int  # extra inlining willingness for non-concrete tuple return types (in hopes of splitting it up)
     inline_error_path_cost::Int # cost of (un-optimized) calls in blocks that throw
 
     # Duplicating for now because optimizer inlining requires it.
@@ -57,7 +57,7 @@ struct OptimizationParams
             inlining::Bool = inlining_enabled(),
             inline_cost_threshold::Int = 100,
             inline_nonleaf_penalty::Int = 1000,
-            inline_tupleret_bonus::Int = 400,
+            inline_tupleret_bonus::Int = 250,
             inline_error_path_cost::Int = 20,
             max_methods::Int = 3,
             tuple_splat::Int = 32,