Fix publish order: codegen must precede cuda-codegen and wgpu-codegen

After Phase 4 DSL consolidation, both cuda-codegen and wgpu-codegen depend
on ringkernel-codegen for shared marker functions. The previous publish
script had them in Tier 1 (no internal deps) but they now belong in Tier 3
(after codegen). Updated tier ordering and header comments.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: mivertowski

scripts/publish.sh

@@ -15,10 +15,11 @@
 #   ./scripts/publish.sh --status            # Check which crates are published
 #
 # The publishing order respects the dependency graph:
-#   Tier 1 (no deps):    core, cuda-codegen, wgpu-codegen, ir
-#   Tier 2 (core deps):  derive, cpu, cuda, wgpu, metal, codegen, ecosystem, audio-fft, graph, montecarlo, cli
-#   Tier 3 (main crate): ringkernel
-#   Tier 4 (apps):       wavesim, txmon, accnet, procint, wavesim3d
+#   Tier 1 (no deps):       core, ir
+#   Tier 2 (core deps):     codegen, cpu, cuda, wgpu, metal, ecosystem, audio-fft, graph, montecarlo
+#   Tier 3 (codegen deps):  cuda-codegen, wgpu-codegen, derive, cli
+#   Tier 4 (main crate):    ringkernel
+#   Tier 5 (apps):          wavesim, txmon, accnet, procint, wavesim3d
 #
 
 set -e
@@ -94,39 +95,39 @@ done
 CRATES=(
     # Tier 1: No internal dependencies
     "ringkernel-core"
-    "ringkernel-cuda-codegen"
-    "ringkernel-wgpu-codegen"
     "ringkernel-ir"
 
-    # Tier 2: Depends only on Tier 1
-    "ringkernel-derive"      # depends on: core, cuda-codegen (optional)
+    # Tier 2: Depends only on core
+    "ringkernel-codegen"     # depends on: core
     "ringkernel-cpu"         # depends on: core
     "ringkernel-cuda"        # depends on: core
     "ringkernel-wgpu"        # depends on: core
     "ringkernel-metal"       # depends on: core
-    "ringkernel-codegen"     # depends on: core
     "ringkernel-ecosystem"   # depends on: core
     "ringkernel-audio-fft"   # depends on: core
     "ringkernel-graph"       # depends on: core
     "ringkernel-montecarlo"  # depends on: core
-    "ringkernel-cli"         # depends on: ir, cuda-codegen (optional), wgpu-codegen (optional)
 
-    # Tier 3: Main crate (depends on most others)
+    # Tier 3: Depends on codegen or other Tier 2 crates
+    "ringkernel-cuda-codegen"  # depends on: codegen
+    "ringkernel-wgpu-codegen"  # depends on: codegen
+    "ringkernel-derive"        # depends on: core, cuda-codegen (optional)
+    "ringkernel-cli"           # depends on: ir, cuda-codegen (optional), wgpu-codegen (optional)
+
+    # Tier 4: Main crate (depends on most others)
     "ringkernel"             # depends on: core, derive, cpu, cuda, wgpu, metal, codegen
 
-    # Tier 4: Application crates (depend on main crate)
+    # Tier 5: Application crates (depend on main crate)
     "ringkernel-wavesim"     # depends on: ringkernel, core, derive, cuda, cuda-codegen
     "ringkernel-txmon"       # depends on: ringkernel, core, cuda, cuda-codegen
-    "ringkernel-accnet"      # depends on: ringkernel, core, cuda, cuda-codegen
-    "ringkernel-procint"     # depends on: ringkernel, core, cuda, cuda-codegen
+    "ringkernel-accnet"      # depends on: core, derive, cpu, cuda-codegen
+    "ringkernel-procint"     # depends on: core, derive, cpu, cuda-codegen
     "ringkernel-wavesim3d"   # depends on: ringkernel, core, derive, cuda, cuda-codegen
 )
 
-# Tier 1 crates can be verified independently
+# Tier 1 crates can be verified independently (no internal deps)
 TIER1_CRATES=(
     "ringkernel-core"
-    "ringkernel-cuda-codegen"
-    "ringkernel-wgpu-codegen"
     "ringkernel-ir"
 )