fix: prevent frontend load errors from leaking in triton-to-cute bridge and add optimizer benchmark baselines

Author: apstenku123

AGENTS.md

@@ -0,0 +1,20 @@
+# AGENTS.md — cppmega.mlx agent contract
+
+## Tensor memory rule
+
+- Wrappers/adapters must not silently allocate or copy large tensors.
+- If a design appears to require staging a large tensor, casting a large tensor,
+  or copying a large tensor only to satisfy a wrapper boundary, treat that
+  design as wrong and keep looking.
+- Prefer passing references to existing GPU buffers, fusion, views/broadcasts,
+  or IR-level lowering through the TileLang/TVM/tvm-ffi pipeline.
+- If zero-copy/fused lowering is not currently possible, fail explicitly or
+  fall back to the existing production path without pretending Path C handled it.
+- Dtype/shape mismatches must be solved at the right level:
+  - first look higher in the graph and make the producer create the tensor in
+    the required dtype/shape from the start;
+  - if the current dtype/layout is the right system-level format (for example
+    FP8 for the active config), move lower and teach the kernel to consume that
+    format directly;
+  - only use casts/reshapes as explicit graph decisions, not hidden adapter
+    staging.