fix(gpu): document enable_ebpf_probes in the system-probe config

With the probes off by default, the gpu_monitoring docs were wrong on two
counts: they said enabling the module loads the eBPF programs, and they never
mentioned the flag that now controls them. A user following the deprecation
note could not find enable_ebpf_probes in their config file, because the
setting was not public and so never rendered into the template.

- describe the module in terms of the NVML metrics it always collects
- expose enable_ebpf_probes as a public setting, marked deprecated, so it
  appears in the generated system-probe.yaml

pkg/config/system-probe_template.yaml is the source the schema enrichment
reads (schema.generate -> parse_template), and check_config_templates.sh
asserts both render identically, so the two files move together.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Author: mbertrone

pkg/config/schema/yaml/system-probe_schema.yaml

@@ -1273,8 +1273,8 @@ properties:
     type: object
     visibility: public
     description: |-
-      Configuration for the GPU Monitoring module, which uses eBPF probes to collect
-      per-process GPU metrics on Linux hosts (kernel 5.8+).
+      Configuration for the GPU Monitoring module, which collects per-process GPU
+      metrics on Linux hosts.
       Required for GPU monitoring on non-containerized Linux hosts alongside
       gpu.enabled: true in datadog.yaml.
     tags: []
@@ -1285,8 +1285,8 @@ properties:
         default: false
         visibility: public
         description: |-
-          Set to true to enable the GPU Monitoring module. This loads the eBPF programs
-          that collect per-process GPU utilization, memory, and performance metrics.
+          Set to true to enable the GPU Monitoring module, which collects per-process
+          GPU utilization, memory, and performance metrics through NVML.
           Must be used together with gpu.enabled: true in datadog.yaml.
       attacher_detailed_logs:
         node_type: setting
@@ -1318,6 +1318,11 @@ properties:
         node_type: setting
         type: boolean
         default: false
+        visibility: public
+        description: |-
+          Set to true to additionally load the GPU Monitoring eBPF probes, which collect
+          the gpu.process.core.usage metric. Requires Linux kernel 5.8 or later.
+          These probes are deprecated and are disabled by default.
       enable_fatbin_parsing:
         node_type: setting
         type: boolean

pkg/config/system-probe_template.yaml

@@ -242,21 +242,29 @@
 ########################################
 
 ## @param gpu_monitoring - custom object - optional
-## Configuration for the GPU Monitoring module, which uses eBPF probes to collect
-## per-process GPU metrics on Linux hosts (kernel 5.8+).
+## Configuration for the GPU Monitoring module, which collects per-process GPU
+## metrics on Linux hosts.
 ## Required for GPU monitoring on non-containerized Linux hosts alongside
 ## gpu.enabled: true in datadog.yaml.
 #
 # gpu_monitoring:
 
 #   # @param enabled - boolean - optional - default: false
 #   # @env DD_GPU_MONITORING_ENABLED - boolean - optional - default: false
-#   # Set to true to enable the GPU Monitoring module. This loads the eBPF programs
-#   # that collect per-process GPU utilization, memory, and performance metrics.
+#   # Set to true to enable the GPU Monitoring module, which collects per-process
+#   # GPU utilization, memory, and performance metrics through NVML.
 #   # Must be used together with gpu.enabled: true in datadog.yaml.
 #
 #   enabled: false
 
+#   # @param enable_ebpf_probes - boolean - optional - default: false
+#   # @env DD_GPU_MONITORING_ENABLE_EBPF_PROBES - boolean - optional - default: false
+#   # Set to true to additionally load the GPU Monitoring eBPF probes, which collect
+#   # the gpu.process.core.usage metric. Requires Linux kernel 5.8 or later.
+#   # These probes are deprecated and are disabled by default.
+#
+#   enable_ebpf_probes: false
+
 {{- if (eq .OS "windows") }}
 
 ##################################################