Remove the dead facades and give the rollback tail an owner

[dhsohn]

Motivation

Three kinds of code that reads like it does something and does not: forwarding
methods whose only callers are the tests that verify the forwarding, a helper
that calls itself through two hops of adapter indirection, and one eight-line
rollback tail copy-pasted five times because nothing owned it. Plus 49 lines of
scene-rect code living in the wrong module of its own package.

Changes

Twenty-three dead facade methods removed, net −488 lines. Each candidate was
re-measured on the current tree rather than taken from the survey: forwarding
methods collected by AST, then production callers counted unfiltered from two
angles — call sites and bare-word occurrences, because six candidates turned out
to be alive only as bound-method references (kwarg=self.method) that a
call-site grep alone would have missed, and those six stay. Removals cascade:
deleting a forwarder often left its target at zero production callers too, so
the unit is the chain, not the method — bond_merge_seed through two layers
down to the module function that has a real internal caller and survives,
get_name_from_smiles down to the 18-entry _name_map dict that existed only
to serve it. Tests that verified only the deleted wiring were deleted with it —
one asserted against a bare mock.Mock() and could never have failed for a real
reason — and tests that verified behavior through a facade were re-aimed at the
real path.

The set_text_align name stays in the boundary file's re-introduction ban list,
which is the ban working as intended, not a leftover.

The scene-rect snapshot trio moved home. capture/release/ restore_scene_rect_snapshot lived in scene_runtime.py — the package's largest
module — while scene_rect.py is the declared owner of everything scene-rect.
Same names, same package, both files already inside the rollback-kernel checks'
glob.

Two adapter round trips flattened. RDKitConversionHelper called two of its
own methods as self.adapter.X(...), which the adapter forwards straight back
to the helper. Read before flattening: those two adapter methods are pure
forwarding with no last_error reset and no RDKitResult wrapping, and no test
patches them on the adapter. The _embed_3d_molecule hops that look similar are
not flattened — the adapter's own comment calls them a seam, and tests
really do patch that seam.

The rollback tail got an owner. Re-measured after the note-consolidation
series, the eight-line "restore the absolute scene/runtime snapshots" tail was
still copied verbatim in five places across history_commands and
scene_runtime. It is now restore_absolute_snapshots() in scene_runtime,
next to the restore machinery it drives; the raise decisions stay at the call
sites.

One thing found and deliberately not done: with its controller forwarder gone,
InsertSmilesService.load_smiles — the whole document-replacing SMILES load
with its exact-rollback machinery — now has zero production callers. That is a
feature removal, not a cleanup, so it is recorded here and left for a separate
decision rather than folded into this diff.

Verification

bash scripts/check.sh — exit 0: ruff clean, formatted, mypy clean, all 300
test files passing in their own processes. The nineteen affected test files pass
directly — 694 tests — including test_core_rdkit_adapter under a real RDKit
install, which covers the flattened calls.

Every "zero consumers" claim in this diff is backed by an unfiltered two-angle
search recorded before the removal, and the survivors list shows the method that
each angle saved. No file mode changed.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-08-30T06:28:18.772910Z	7f6e749	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.