[Codegen] Re-Enable transform dialect configuration strategy round 2 (#16427)

This time it just drops all transform dialect usage outside of transform
library file path + entry point name. This reduces code complexity in
`MaterializeUserConfigs`.

Also cleans up some of the transform dialect tests to stop lit testing
at the same time. We might want to consider dropping some of them as
they aren't being maintained (the only thing they verify at the moment
is that the transform scripts are valid for CUDA).
diff --git a/samples/transform_dialect/example_module.mlir b/samples/transform_dialect/example_module.mlir
index c5eab46..1e4ac4e 100644
--- a/samples/transform_dialect/example_module.mlir
+++ b/samples/transform_dialect/example_module.mlir
@@ -107,28 +107,29 @@
 }
 
 /// We test first with threading off so that the printers are legible.
-// R-UN: iree-compile %s --iree-hal-target-backends=vulkan \
-// R-UN:   --iree-codegen-use-transform-dialect-strategy=transform_main \
-// R-UN:   --iree-codegen-transform-dialect-library=%p/transform_library.mlir \
-// R-UN:   --compile-from=executable-sources \
-// R-UN:   --compile-to=executable-targets \
-// R-UN:   --mlir-disable-threading | \
-// R-UN: FileCheck %s --check-prefixes=CODEGEN-PRINTER
+// RUN: iree-compile %s --iree-hal-target-backends=vulkan \
+// RUN:   --iree-codegen-transform-dialect-library=%p/transform_library.mlir@kernel_config \
+// RUN:   --compile-from=executable-sources \
+// RUN:   --compile-to=executable-targets \
+// RUN:   --mlir-disable-threading | \
+// RUN: FileCheck %s --check-prefixes=CODEGEN-PRINTER
 
-// CODEGEN-PRINTER:     IR printer: Setting matmul strategy to default top-level
-// CODEGEN-PRINTER:       translation_info = #iree_codegen.translation_info<TransformDialectCodegen codegen_spec = @transform_main
+// CODEGEN-PRINTER:     IR printer: Setting matmul strategy to custom_transform_strategy
+// CODEGEN-PRINTER:       translation_info = #iree_codegen.translation_info<TransformDialectCodegen codegen_spec = @custom_transform_strategy>
 // CODEGEN-PRINTER:     IR printer: Setting reduce strategy to base vectorize top-level
 // CODEGEN-PRINTER:       translation_info = #iree_codegen.translation_info<SPIRVBaseVectorize>, workgroup_size = [16 : index, 1 : index, 1 : index]
 
 /// Then test with threading to make sure it runs
 // RUN: iree-compile %s --iree-hal-target-backends=vulkan \
-// RUN:   --iree-codegen-use-transform-dialect-strategy=@transform_main \
-// RUN:   --iree-codegen-transform-dialect-library=%p/transform_library.mlir \
+// RUN:   --iree-codegen-transform-dialect-library=%p/transform_library.mlir@kernel_config \
 // RUN:   --compile-from=executable-sources \
 // RUN:   --compile-to=executable-targets \
 // RUN:   --mlir-disable-threading | \
 // RUN: FileCheck %s --check-prefixes=CODEGEN
 
+// CODEGEN: Ran custom_transform_strategy
 // CODEGEN: spirv.func @example_module_dispatch_0_generic_80_f32
-// CODEGEN: spirv.func @example_module_dispatch_1_matmul_16x16x5_f32
+// CODEGEN: hal.executable private @example_module_dispatch_1
+// CODEGEN:   #iree_codegen.translation_info<TransformDialectCodegen codegen_spec = @custom_transform_strategy>
+// CODEGEN:     spirv.func @example_module_dispatch_1_matmul_16x16x5_f32
 // CODEGEN: spirv.func @example_module_dispatch_2_generic_16x16_f32