Fix procedure repr regression: restore #<procedure> display

PR #43's perf optimization switched procedures from cons-list to raw
Python tuples, but nothing gave the tuple a custom repr, so notebook
Out[] cells and Python repr() dumped the raw (tag, fn, bodies, formals,
env, safe) contents instead of "#<procedure>". make_proc now wraps its
result in _ProcTuple, a zero-overhead tuple subclass with __repr__,
leaving the hotter continuation/fail/handler construction paths
untouched.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Author: dsblank

calysto_scheme/scheme.py

@@ -545,8 +545,16 @@ def vector_length(vec):
 ### literals instead of these names -- see the comment on _eval_direct
 ### about extra-overhead-per-node being measurable on those paths.
 
+class _ProcTuple(tuple):
+    # A plain tuple subclass so procedures still repr as "#<procedure>"
+    # (matching the old cons-based representation) instead of dumping
+    # their raw (fn, bodies, formals, env, safe) contents.
+    __slots__ = ()
+    def __repr__(self):
+        return "#<procedure>"
+
 def make_proc(*args):
-    return (symbol_procedure,) + args
+    return _ProcTuple((symbol_procedure,) + args)
 
 def make_macro(*args):
     return (symbol_macro_transformer,) + args

calysto_scheme/src/Scheme.py

@@ -537,8 +537,16 @@ def vector_length(vec):
 ### literals instead of these names -- see the comment on _eval_direct
 ### about extra-overhead-per-node being measurable on those paths.
 
+class _ProcTuple(tuple):
+    # A plain tuple subclass so procedures still repr as "#<procedure>"
+    # (matching the old cons-based representation) instead of dumping
+    # their raw (fn, bodies, formals, env, safe) contents.
+    __slots__ = ()
+    def __repr__(self):
+        return "#<procedure>"
+
 def make_proc(*args):
-    return (symbol_procedure,) + args
+    return _ProcTuple((symbol_procedure,) + args)
 
 def make_macro(*args):
     return (symbol_macro_transformer,) + args