infra(kapsule): remove h100x2 + h100-sxm pools

Reasons:
- h100-sxm: Scaleway account quota is 0/0 for the SXM instance type
  (cp_servers_type_H100_SXM_2_80G). The pool was perpetually trying
  to maintain size=1 with a creation_error node, which JAMMED the
  cluster autoscaler. That blocked L40S scale-up entirely during
  today's alpha-perception cluster run.
- h100x2: more expensive than current workloads justify. Every
  training path (alpha perception, future PPO) fits on either the
  single-GPU H100 or L40S.

Removed:
- Two `scaleway_k8s_pool` resources from infra/modules/kapsule/main.tf
- Six variables (enable + type + max_size for each pool)
- Two outputs (pool_id for each)
- Corresponding inputs in infra/live/production/kapsule/terragrunt.hcl

The live cluster has the SXM pool stuck node manually deleted via
`scw k8s node delete` (this commit-session); the pool resource itself
will be destroyed on next `terragrunt apply`.

Post-cleanup pool inventory:
- platform (DEV1-L × 3)
- ci-training-h100  (H100-1-80G, max 1)  <- regular single-GPU
- ci-training-l40s  (L40S-1-48G, max 1)  <- primary training target
- ci-compile-cpu    (POP2-HC, max 4)
- ci-compile-cpu-hm (POP2-HM, max 1)

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

infra/live/production/kapsule/terragrunt.hcl

@@ -38,15 +38,11 @@ inputs = {
   ci_training_h100_type        = "H100-1-80G"
   ci_training_h100_max_size    = 1
 
-  # H100x2 training pool (2x 80GB — multi-GPU parallel hyperopt)
+  # H100x2 and H100-SXM pools removed 2026-05-16:
-  enable_ci_training_h100x2_pool = true
+  # - SXM: Scaleway account quota is 0/0 (cp_servers_type_H100_SXM_2_80G).
-  ci_training_h100x2_type        = "H100-2-80G"
+  #        The pool was stuck trying to provision and jammed the autoscaler.
-  ci_training_h100x2_max_size    = 1
+  # - H100x2: more expensive than current workloads justify; the single-GPU
-
+  #          H100 pool covers all training paths.
-  # H100-SXM training pool (2x 80GB SXM, NVLink)
-  enable_ci_training_h100_sxm_pool = true
-  ci_training_h100_sxm_type        = "H100-SXM-2-80G"
-  ci_training_h100_sxm_max_size    = 1
 
   # NAT gateway handles egress — no public IPs on nodes
   public_ip_disabled = true

infra/modules/kapsule/main.tf

@@ -108,43 +108,12 @@ resource "scaleway_k8s_pool" "ci_training_h100" {
   }
 }
 
-# H100x2 training pool (2x 80GB VRAM — multi-GPU parallel hyperopt)
+# NOTE: ci-training-h100x2 and ci-training-h100-sxm pools removed
-resource "scaleway_k8s_pool" "ci_training_h100x2" {
+# 2026-05-16 — Scaleway account doesn't have quota for the SXM variant
-  count       = var.enable_ci_training_h100x2_pool ? 1 : 0
+# (cp_servers_type_H100_SXM_2_80G 0/0), and the H100x2 variant is
-  cluster_id  = scaleway_k8s_cluster.foxhunt.id
+# more expensive than our actual usage justifies. Keeping the regular
-  name        = "ci-training-h100x2"
+# single-GPU H100 pool (ci_training_h100) + the L40S pool covers
-  node_type   = var.ci_training_h100x2_type
+# every current training workload.
-  size        = 1
-  min_size    = 0
-  max_size    = var.ci_training_h100x2_max_size
-  autoscaling        = true
-  autohealing        = true
-  public_ip_disabled = var.public_ip_disabled
-  region             = var.region
-
-  lifecycle {
-    ignore_changes = [size]
-  }
-}
-
-# H100-SXM training pool (2x 80GB SXM VRAM — NVLink multi-GPU)
-resource "scaleway_k8s_pool" "ci_training_h100_sxm" {
-  count       = var.enable_ci_training_h100_sxm_pool ? 1 : 0
-  cluster_id  = scaleway_k8s_cluster.foxhunt.id
-  name        = "ci-training-h100-sxm"
-  node_type   = var.ci_training_h100_sxm_type
-  size        = 1
-  min_size    = 0
-  max_size    = var.ci_training_h100_sxm_max_size
-  autoscaling        = true
-  autohealing        = true
-  public_ip_disabled = var.public_ip_disabled
-  region             = var.region
-
-  lifecycle {
-    ignore_changes = [size]
-  }
-}
 
 # CPU compile pool (POP2-HC — high CPU, no GPU, for cargo check/build)
 resource "scaleway_k8s_pool" "ci_compile_cpu" {

infra/modules/kapsule/outputs.tf

@@ -34,16 +34,6 @@ output "ci_training_h100_pool_id" {
   value       = var.enable_ci_training_h100_pool ? scaleway_k8s_pool.ci_training_h100[0].id : ""
 }
 
-output "ci_training_h100x2_pool_id" {
-  description = "ID of the H100x2 training node pool"
-  value       = var.enable_ci_training_h100x2_pool ? scaleway_k8s_pool.ci_training_h100x2[0].id : ""
-}
-
-output "ci_training_h100_sxm_pool_id" {
-  description = "ID of the H100-SXM training node pool"
-  value       = var.enable_ci_training_h100_sxm_pool ? scaleway_k8s_pool.ci_training_h100_sxm[0].id : ""
-}
-
 output "private_network_id" {
   description = "ID of the VPC private network the cluster is attached to"
   value       = scaleway_vpc_private_network.foxhunt.id

infra/modules/kapsule/variables.tf

@@ -59,42 +59,6 @@ variable "ci_training_h100_max_size" {
   default     = 1
 }
 
-variable "enable_ci_training_h100x2_pool" {
-  description = "Create H100x2 GPU node pool for multi-GPU training"
-  type        = bool
-  default     = false
-}
-
-variable "ci_training_h100x2_type" {
-  description = "Instance type for the H100x2 training pool"
-  type        = string
-  default     = "H100-2-80G"
-}
-
-variable "ci_training_h100x2_max_size" {
-  description = "Maximum number of nodes in the H100x2 training pool"
-  type        = number
-  default     = 1
-}
-
-variable "enable_ci_training_h100_sxm_pool" {
-  description = "Create H100-SXM GPU node pool for NVLink multi-GPU training"
-  type        = bool
-  default     = false
-}
-
-variable "ci_training_h100_sxm_type" {
-  description = "Instance type for the H100-SXM training pool"
-  type        = string
-  default     = "H100-SXM-2-80G"
-}
-
-variable "ci_training_h100_sxm_max_size" {
-  description = "Maximum number of nodes in the H100-SXM training pool"
-  type        = number
-  default     = 1
-}
-
 variable "enable_ci_compile_cpu_pool" {
   description = "Create high-CPU node pool for CI compilation"
   type        = bool