ci: cache elc binary + Docker layers, asset changes from 42min to ~5min

- deploy.yaml: restore elc from GCS (gs://neuron-ci-cache) keyed on
  source SHA; only compile on cache miss, then upload for future runs
- Dockerfile.stage: pre-compile el_runtime.o as its own layer so the
  expensive object is cached when only main.c changes between runs
- build-stage.sh: add --cache-from/--cache-to pointing at Artifact
  Registry so apt-get + compilation layers survive across cold builds

Dockerfile.stage

@@ -25,6 +25,11 @@ WORKDIR /build
 # El runtime (shared by both binaries)
 COPY runtime/el_runtime.c runtime/el_runtime.h ./
 
+# Pre-compile el_runtime as a separate cached layer.
+# el_runtime.c changes rarely; main.c changes every run.
+# Splitting this out means el_runtime.o is cached across builds when only main.c changes.
+RUN cc -O2 -c el_runtime.c -I. -o el_runtime.o
+
 # ── Build neuron-web ──────────────────────────────────────────────────────────
 #
 # main.c was generated on the host by build-stage.sh from src/*.el via the
@@ -35,7 +40,7 @@ COPY dist/main.c ./
 
 RUN cc -O2 -rdynamic \
        -o neuron-web \
-       main.c web_stubs.c el_runtime.c \
+       main.c web_stubs.c el_runtime.o \
        -lcurl -lpthread -ldl -lm -lssl -lcrypto
 
 # ── Build soul-demo ───────────────────────────────────────────────────────────
@@ -44,7 +49,7 @@ COPY dist/vessel_stubs.c ./
 
 RUN cc -O2 -rdynamic \
        -o soul-demo \
-       soul-demo.c vessel_stubs.c el_runtime.c \
+       soul-demo.c vessel_stubs.c el_runtime.o \
        -lcurl -lpthread -ldl -lm -lssl -lcrypto
 
 # ── Stage 2: runtime image ────────────────────────────────────────────────────