Update on "[ExecuTorch] Make ForcedUnroll usage in bf16 BlasKernel actually work for -Oz builds"

Clang is very resistant to inlining under -Oz. For ForcedUnroll to actually unroll, we need to force-inline the lambda.

Differential Revision: [D62154247](https://our.internmc.facebook.com/intern/diff/D62154247/)

[ghstack-poisoned]

Author: swolchok

build/cmake_deps.toml

@@ -116,6 +116,20 @@ deps = [
   "executorch",
 ]
 
+[targets.optimized_native_cpu_ops_oss]
+buck_targets = [
+  "//configurations:optimized_native_cpu_ops_oss",
+]
+filters = [
+  ".cpp$",
+]
+excludes = [
+]
+deps = [
+  "executorch_no_prim_ops",
+  "executorch",
+  "portable_kernels",
+]
 # ---------------------------------- core end ----------------------------------
 # ---------------------------------- extension start ----------------------------------
 [targets.extension_data_loader]
@@ -341,5 +355,6 @@ deps = [
   "portable_kernels",
   "quantized_kernels",
   "xnnpack_backend",
+  "optimized_native_cpu_ops_oss",
 ]
 # ---------------------------------- LLama end ----------------------------------