Enable MLGO inliner optimization for optimize_for_size cases

This helps reduce binary size for these projects by ~3%.

Test: presubmit
Bug: 342090838
Change-Id: Ie1e0586ddf5f40aa1e81fc2628a6499093de9699
diff --git a/cc/lto.go b/cc/lto.go
index a084db7..60eb4d6 100644
--- a/cc/lto.go
+++ b/cc/lto.go
@@ -144,7 +144,7 @@
 
 		if !ctx.Config().IsEnvFalse("THINLTO_USE_MLGO") {
 			// Register allocation MLGO flags for ARM64.
-			if ctx.Arch().ArchType == android.Arm64 {
+			if ctx.Arch().ArchType == android.Arm64 && !ctx.optimizeForSize() {
 				ltoLdFlags = append(ltoLdFlags, "-Wl,-mllvm,-regalloc-enable-advisor=release")
 			}
 			// Flags for training MLGO model.