fix(fly): cold-start hardening — 90s grace, HTTP health check, curl

- fly.toml: switch [[services.tcp_checks]] → [[services.http_checks]] on
  /api/health so Fly tests the server, not just the open port; bump
  grace_period 30s → 90s (Isabelle Main heap + Julia JIT need the headroom)
- .containerization/Containerfile.full: add curl to runtime apt-get so the
  HEALTHCHECK can use it; change HEALTHCHECK from --version (binary exists
  check) to curl -fs http://localhost:8090/api/health; start-period 5s → 90s
- Comments: .internal → .flycast correction (auto-wake requires flycast)

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

Author: hyperpolymath

.containerization/Containerfile.full

@@ -188,13 +188,15 @@ LABEL org.opencontainers.image.vendor="hyperpolymath"
 
 # Runtime dependencies only.
 # `perl` + `fontconfig` are Isabelle's launcher-script requirements.
+# `curl` is needed for the HEALTHCHECK below.
 RUN apt-get update && apt-get install -y --no-install-recommends \
     ca-certificates \
     libssl3 \
     libgmp10 \
     perl \
     fontconfig \
     z3 \
+    curl \
     && rm -rf /var/lib/apt/lists/*
 
 WORKDIR /app
@@ -240,8 +242,9 @@ EXPOSE 50051
 # Julia ML
 EXPOSE 8090
 
-HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
-    CMD ["/app/bin/echidna", "--version"]
+# start-period=90s: Isabelle Main heap (~1-2 GB) + Julia JIT take time on cold start.
+HEALTHCHECK --interval=30s --timeout=10s --start-period=90s --retries=3 \
+    CMD curl -fs http://localhost:8090/api/health || exit 1
 
 ENTRYPOINT ["/app/bin/echidna"]
 CMD ["--help"]

fly.toml

@@ -5,8 +5,9 @@
 #   flyctl launch --no-deploy --copy-config --name echidna-nesy
 #   flyctl deploy -c fly.toml -f .containerization/Containerfile.full
 #
-# Internal-only: echidna is reached via echidna-nesy.internal:8090 from
+# Internal-only: echidna is reached via echidna-nesy.flycast:8090 from
 # proven-nesy-solver-api; no public http_service.
+# Use .flycast (not .internal) so Fly auto-wakes the machine on first connection.
 #
 # NB: Container is LARGE (~6-8 GB) — Julia + Idris2 + Agda + Lean + Isabelle.
 # Memory must be >=4GB (Isabelle Main heap ~1GB); shared-cpu-2x is the
@@ -43,10 +44,15 @@ kill_timeout = "10s"
     port = 8090
     handlers = []
 
-  [[services.tcp_checks]]
+  # HTTP check on /api/health — tests the server is serving, not just the port.
+  # grace_period must be generous: Isabelle Main heap maps ~1-2 GB on first load
+  # and Julia has JIT overhead on first prover call.  90s is the minimum safe value.
+  [[services.http_checks]]
     interval = "30s"
-    timeout = "5s"
-    grace_period = "30s"
+    timeout = "10s"
+    grace_period = "90s"
+    method = "GET"
+    path = "/api/health"
 
 [[vm]]
   size = "shared-cpu-2x"