optimization_guide: flag missing graph-stats, fix walkthrough citation

The doc's primary diagnostic (`make graph-stats`) is referenced six
times throughout but the Make target was removed in Step E (it shipped
with the DOOM renderer). The principles in the doc are general — they
apply to any compiled graph — but the operational instructions point
at deleted tooling.

Two changes:

- Added a top-of-doc note explaining that `make graph-stats`
  references describe a removed diagnostic, with a pointer to where
  the original implementation lives. The DOOM-anchored worked
  examples in the body are now flagged as illustrative — they
  preserve the original numbers that informed each rule of thumb.

- Section § 11 'Correctness checks after structural changes' cited
  `make walkthrough` (removed) and 'the relevant test file under
  tests/doom/' (deleted directory) as the authoritative multi-position
  check. Replaced with `make test`, which is the actual surviving
  authoritative check.

The body's `make graph-stats` mentions are intentionally left in
place — they describe a real diagnostic concept ('per-annotation costs
and layer counts'), and the top-of-doc note now scopes them as
historical references rather than actionable instructions.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: physicsrob

optimization_guide.md

@@ -6,10 +6,15 @@ compiler are small and readable (`torchwright/compiler/forward/`);
 read the source when a detail matters.
 
 This guide is deliberately light on specific numbers — the graph and
-compiler evolve, and quantitative snapshots go stale. Run
-`make graph-stats` for current per-annotation costs, layer counts,
-and critical-path targets; this doc is for the concepts behind those
-numbers.
+compiler evolve, and quantitative snapshots go stale. References below
+to `make graph-stats` describe a per-annotation/per-stage diagnostic
+that was removed from this repo in Step E (it shipped together with
+the DOOM renderer and now lives in the sibling `torchwright_doom`
+repo). The principles still apply to any compiled graph; replicate
+the diagnostic from the verbose `compile_headless` output, or pull the
+original implementation from the sibling repo if you need it. The
+DOOM-anchored worked examples in this doc are illustrative — they
+preserve the original numbers that informed each rule of thumb.
 
 ---
 
@@ -580,9 +585,8 @@ layers indicate a wide intermediate living too long.
 `torchwright/debug/probe.py` runs the compiled module side-by-side
 with a recursive oracle evaluator for a single position and reports
 the first divergence. Run it after any graph restructuring. For
-multi-position / autoregressive behaviour, the golden-frame test
-(`make walkthrough`) or the relevant test file under `tests/doom/`
-is the authoritative check.
+multi-position / autoregressive behaviour, the test suite (`make
+test`) is the authoritative check.
 
 ### Attribute layer count to a subsystem