Make constant_folding's _DEFAULT_SKIP_TARGETS public (#10760)

Summary:

This follows D73513516 and allows customers of constant_prop_pass to use previous defaults

Differential Revision: D74349918

Author: ThomasJannaud

exir/passes/constant_prop_pass.py

@@ -29,7 +29,8 @@
 
 # Avoid propagating constants for `exir.ops.edge.aten.full.default`.
 # Propagating aten.full can significantly increase compiled model size.
-_DEFAULT_SKIP_TARGETS = {exir_ops.edge.aten.full.default}
+_DEFAULT_SKIP_TARGETS_NO_QUANT = {exir_ops.edge.aten.full.default}
+_DEFAULT_SKIP_TARGETS = set(_DEFAULT_SKIP_TARGETS_NO_QUANT)
 
 # Do not const prop quantization primitives
 _QUANT_PRIMITIVES_EDGE = [aten_to_edge(op) for op in _QUANT_PRIMITIVES]
@@ -47,6 +48,9 @@
     torch.layout,
 )
 
+def get_default_skip_targets_no_quant() -> set[EdgeOpOverload]:
+    return _DEFAULT_SKIP_TARGETS_NO_QUANT
+
 
 def is_const(
     arg,