analytical-viz: batch B in every formula string + rename 'latency-optimal' → 'memory-optimal'

Two related fixes.

1) Formula strings show B alongside every other symbol. Previously the
   numeric FLOPs/mem_bytes/comm_bytes were scaled by B, but the printed
   formula strings still read like B=1 amounts. That made the per-stage
   table confusing at high B: numbers moved but formulas didn't. Updated
   every attention + FFN stage's `formula` / `flops_formula` /
   `mem_formula` / `comm_formula` to include B explicitly. Weight-bytes
   lines now say "(shared across batch)" / "(weight, B-invariant)" so it's
   obvious what does and doesn't scale.

   Stages touched: S1..S10 (attention), C1/C2/C3 (attn comm), F1..F5
   (FFN), CF1 (FFN AR).

2) Renamed the button/spinner/help copy from "latency-optimal" to
   "memory-optimal" to match the smallest-fit pick semantics I put in
   place earlier. The buttons already picked smallest-fit; the labels
   just still said "latency-optimal" from the previous iteration.

   Left the Auto Hardware sensitivity chart's "latency-optimal baseline"
   label alone — that panel is a HW co-design view where "how fast can
   each HW go?" is the intended question.

Verified: 24 pytest tests still pass.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

tests/analytical_visualization/app.py

@@ -218,8 +218,8 @@ with st.sidebar.expander("Parallelism", expanded=True):
         st.session_state["cp_placement"] = _default_suggestion.cp_placement
         st.session_state["_pl_active_scope"] = "full"
         st.rerun()
-    # Three latency-optimal apply buttons — pick a scope, load the
-    # latency-min Pareto config into the sidebar sliders.
+    # Three memory-optimal apply buttons — pick a scope, load the
+    # smallest-fit Pareto config into the sidebar sliders.
     from tests.analytical_visualization.auto_explore import (
         run_auto_explore as _run_auto_explore_sb,
     )
@@ -235,14 +235,14 @@ with st.sidebar.expander("Parallelism", expanded=True):
             _sb_clicks[_sb_scope] = st.button(
                 _sb_scope_meta[_sb_scope][2], width='stretch',
                 key=f"_sb_apply_{_sb_scope}",
-                help=f"Apply latency-optimal parallelism for "
+                help=f"Apply memory-optimal (smallest-fit) parallelism for "
                      f"{_sb_scope_meta[_sb_scope][2]} scope.",
             )
     for _sb_scope, _sb_click in _sb_clicks.items():
         if not _sb_click:
             continue
         _sb_ia, _sb_ff, _sb_lbl = _sb_scope_meta[_sb_scope]
-        with st.spinner(f"Finding latency-optimal parallelism ({_sb_lbl})..."):
+        with st.spinner(f"Finding memory-optimal parallelism ({_sb_lbl})..."):
             _sb_res = _run_auto_explore_sb(
                 model, _default_machine, s_kv=s_kv, mode=mode,
                 include_attention=_sb_ia, include_ffn=_sb_ff, b=b_batch,
@@ -553,17 +553,18 @@ tab_layout, tab_auto, tab_memory, tab_stages, tab_compare, tab_hw = st.tabs([
 
 # ── TAB 1: Physical layout ──────────────────────────────────────
 with tab_layout:
-    # Quick shortcut: three buttons that pick the latency-optimal Pareto
-    # config for a chosen scope and load it into the sidebar. Layout
-    # re-renders immediately on the next Streamlit run with the new config.
+    # Quick shortcut: three buttons that pick the memory-optimal
+    # (smallest-fit) Pareto config for a chosen scope and load it into
+    # the sidebar. Layout re-renders immediately on the next Streamlit
+    # run with the new config.
     from tests.analytical_visualization.auto_explore import (
         run_auto_explore as _run_auto_explore_layout,
     )
 
     st.markdown(
-        "**Quick auto-suggest layout** — pick a scope; the latency-optimal "
-        "Pareto config for that scope is loaded into the sidebar and this "
-        "layout redraws around it."
+        "**Quick auto-suggest layout** — pick a scope; the memory-optimal "
+        "(smallest-fit) Pareto config for that scope is loaded into the "
+        "sidebar and this layout redraws around it."
     )
     _pl1, _pl2, _pl3, _pl4 = st.columns([1, 1, 1, 1])
     with _pl1:
@@ -591,7 +592,7 @@ with tab_layout:
         if not _pl_clicked:
             continue
         _pl_ia, _pl_ff, _pl_lbl = _pl_scope_meta[_pl_scope]
-        with st.spinner(f"Finding latency-optimal parallelism ({_pl_lbl})..."):
+        with st.spinner(f"Finding memory-optimal parallelism ({_pl_lbl})..."):
             _pl_res = _run_auto_explore_layout(
                 model, machine, s_kv=s_kv, mode=mode,
                 include_attention=_pl_ia, include_ffn=_pl_ff, b=b_batch,
@@ -629,7 +630,7 @@ with tab_layout:
             # stage tables (attn/ffn/full).
             st.session_state["_pl_active_scope"] = _pl_scope
             st.success(
-                f"Loaded latency-optimal {_pl_lbl} config: CP={_pl_best.cp} "
+                f"Loaded memory-optimal {_pl_lbl} config: CP={_pl_best.cp} "
                 f"TP={_pl_best.tp} PP={_pl_best.pp} DP={_pl_best.dp} → "
                 f"{_pl_best.latency_ms:.2f} ms, {_pl_best.pes_used} PEs. "
                 "Redrawing layout..."