[frontend] Fix autotune cache lookup when interpreter enabled (#6678)

The newly-added autotune cache needs to bail out if given an
InterpretedFunction (it has no cache key, and autotuning the interpreter
is a little meaningless)

Author: bertmaher

python/triton/runtime/autotuner.py

@@ -32,7 +32,7 @@ def __init__(self, fn, arg_names, configs, key, reset_to_zero, restore_value, pr
         self.keys = key
         self.cache: Dict[Tuple, Config] = {}
         self.arg_names = arg_names
-        self.cache_results = cache_results or knobs.autotuning.cache
+        self.cache_results = cache_results or (knobs.autotuning.cache and not knobs.runtime.interpret)
 
         # Reset to zero or restore values
         self.reset_to_zero = []