Indexed dispatch over comprehension-built and non-append-mutated containers has no callees

[khatchad]

Follow-up to the indexed-dispatch shape of #661 (fixed for append-built lists in ponder-lab#533).

The ponder-lab#533 fix surfaces append-accumulated contents at subscript reads by reading the synthetic __list_append_contents__ property (processListAppend's write-side dual). Containers populated any other way still starve an indexed dispatch:

• Comprehension-built (confirmed): replacing the fixture's append loop with self.sub_layers = [Inner() for _ in range(self.num_layers)] fails Function must exist in call graph for Inner.call on the Surface append-accumulated list contents at subscript reads ponder-lab/ML#533 branch—comprehension contents flow through neither the append property nor numeric fields. This form is at least as common as append in Keras subclasses.
• Other mutators (structural, unverified): extend, insert, and += are unmodeled entirely (processListAppend matches append only), so they starve both subscript reads and value iteration.
• Dict-keyed dispatch (unverified): self.blocks["enc"](x) reads a constant-string member, deliberately outside the Surface append-accumulated list contents at subscript reads ponder-lab/ML#533 gate; literal/[k] =-populated dicts likely work but have no guard.

A runnable comprehension fixture (tf2_test_layer_list_compr.py, the tf2_test_layer_list.py miniature with the accumulation swapped) reproduces the first bullet; suggest landing it as a suppressed guard with a TODO referencing this issue, then generalizing content surfacing per accumulation form.