feat(infra): upgrade CI to L40S + mold linker for faster builds

- Switch linker from lld to mold (~2-5x faster linking for large binaries)
  - Install mold 2.35.1 in CI builder Dockerfile
  - Update .cargo/config.toml: -fuse-ld=mold
- Upgrade CI build pool: L4-1-24G → L40S-1-48G (~2x training throughput)
  - Increase max_size from 1 to 2 (allows concurrent jobs, fixes scheduling deadlocks)
  - Update runner resource limits for L40S node (24 vCPU, 96GB)
- Update runner-values.yaml comments and .gitlab-ci.yml header

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

.cargo/config.toml

@@ -26,7 +26,7 @@ progress.when = "auto"
 [target.x86_64-unknown-linux-gnu]
 linker = "clang"
 rustflags = [
-    "-C", "link-arg=-fuse-ld=lld",
+    "-C", "link-arg=-fuse-ld=mold",
     "-C", "link-arg=-Wl,-z,relro,-z,now",
     "-C", "link-arg=-Wl,--as-needed",
     # AVX2/FMA baseline — matches both Scaleway L4/H100 and most dev machines

.gitlab-ci.yml

@@ -1,5 +1,5 @@
 # GitLab CI/CD — Foxhunt
-# Runs on ci-build pool (L4, CUDA 8.9, scale-to-zero), linker: clang+lld
+# Runs on ci-build pool (L40S, CUDA 8.9, autoscales 0-2), linker: clang+mold
 # CI builder image: CUDA 12.4 + Rust 1.89 + protoc + sccache (hosted on Scaleway CR)
 # Service images: pushed to Scaleway Container Registry (Kaniko)
 # Note: .gitlab-ci.yml itself is NOT in changes: filters — CI-only edits skip build

infra/docker/Dockerfile.ci-builder

@@ -23,6 +23,11 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
     unzip \
     && rm -rf /var/lib/apt/lists/*
 
+# mold linker — 2-5x faster than lld for final link step on large Rust binaries
+RUN curl -fsSL https://github.com/rui314/mold/releases/download/v2.35.1/mold-2.35.1-x86_64-linux.tar.gz \
+    | tar xz -C /usr/local --strip-components=1 \
+    && chmod +x /usr/local/bin/mold
+
 # protoc 28.3 (proto3 optional support, matches local dev)
 RUN curl -fsSL https://github.com/protocolbuffers/protobuf/releases/download/v28.3/protoc-28.3-linux-x86_64.zip -o /tmp/protoc.zip \
     && unzip -o /tmp/protoc.zip -d /usr/local bin/protoc 'include/*' \

infra/k8s/gitlab/runner-values.yaml

@@ -1,6 +1,6 @@
-# GitLab Runner — Kubernetes executor targeting ci-build pool (L4)
+# GitLab Runner — Kubernetes executor targeting ci-build pool (L40S)
 # Runner manager pod lives on gitlab node pool
-# Build pods spawn on ci-build pool (L4-1-24G, scale-to-zero)
+# Build pods spawn on ci-build pool (L40S-1-48G, autoscales 0-2)
 
 gitlabUrl: http://gitlab-webservice-default.foxhunt.svc.cluster.local:8181
 # runnerToken set via --set at install time
@@ -26,13 +26,13 @@ runners:
         namespace = "foxhunt"
         image = "rust:1.89-slim"
         privileged = false
-        # L4 scale-to-zero needs ~3-5 min to provision; default 180s times out
+        # L40S scale-to-zero needs ~3-5 min to provision; default 180s times out
         poll_timeout = 600
-        # Resource limits for build pods (L4 node: 12 vCPU, 45.5GB)
-        cpu_request = "4000m"
-        cpu_limit = "10000m"
-        memory_request = "16Gi"
-        memory_limit = "38Gi"
+        # Resource limits for build pods (L40S node: ~24 vCPU, ~96GB)
+        cpu_request = "6000m"
+        cpu_limit = "20000m"
+        memory_request = "24Gi"
+        memory_limit = "80Gi"
         helper_cpu_request = "100m"
         helper_cpu_limit = "500m"
         helper_memory_request = "128Mi"

infra/live/production/kapsule/terragrunt.hcl

@@ -25,10 +25,11 @@ inputs = {
   enable_gitlab_pool = true
   gitlab_type        = "GP1-XS"
 
-  # CI build pool (L4 for cost-effective CI: 12 vCPU, 45.5GB, CUDA CC 89)
+  # CI build pool (L40S for faster CI: 48GB VRAM, ~2x L4 throughput, CUDA CC 89)
+  # max_size=2 allows autoscaler to add a second node for concurrent jobs
   enable_ci_build_pool = true
-  ci_build_type        = "L4-1-24G"
-  ci_build_max_size    = 1
+  ci_build_type        = "L40S-1-48G"
+  ci_build_max_size    = 2
 
   # Dev pool (GP1-L for DevPod remote development, autoscales to zero)
   enable_dev_pool   = true

infra/modules/kapsule/main.tf

@@ -89,7 +89,7 @@ resource "scaleway_k8s_pool" "gitlab" {
   region      = var.region
 }
 
-# CI build pool (L4 — cheaper than H100, has CUDA for compilation + GPU tests)
+# CI build pool (L40S — 48GB VRAM, ~2x L4 throughput, same Ada arch / CUDA CC 89)
 resource "scaleway_k8s_pool" "ci_build" {
   count       = var.enable_ci_build_pool ? 1 : 0
   cluster_id  = scaleway_k8s_cluster.foxhunt.id

infra/modules/kapsule/variables.tf

@@ -76,15 +76,15 @@ variable "enable_ci_build_pool" {
 }
 
 variable "ci_build_type" {
-  description = "Instance type for the CI build node pool"
+  description = "Instance type for the CI build node pool (L40S for fast compilation + training)"
   type        = string
-  default     = "L4-1-24G"
+  default     = "L40S-1-48G"
 }
 
 variable "ci_build_max_size" {
-  description = "Maximum number of nodes in the CI build pool"
+  description = "Maximum number of nodes in the CI build pool (2 allows concurrent jobs)"
   type        = number
-  default     = 1
+  default     = 2
 }
 
 variable "enable_dev_pool" {