Add FOM on the vm_id that executes the job

This is occasionally useful as an inexact way to locate the VM that runs
rank 0.

Sample output:

```
 modifier::GcpMetadata::machine-type = h3-standard-88
    modifier::GcpMetadata::image = projects/schedmd-slurm-public/global/images/slurm-gcp-6-6-hpc-rocky-linux-8-1721332881
    modifier::GcpMetadata::ghostname = caeslurmv6-h3nodeset-0.c.hpc-workload-performance.internal
    modifier::GcpMetadata::main-gid = 6390746155122030581
    modifier::GcpMetadata::gids = 4685835775755810805, 6390746155122030581
    modifier::GcpMetadata::project-id = 767663353010
    modifier::GcpMetadata::Level 0 Groups (cluster) = 1
    modifier::GcpMetadata::Level 1 Groups (rack) = 1
    modifier::GcpMetadata::Level 2 Groups (host) = 2
    modifier::GcpMetadata::All Hosts = e2bf4b97dbb6b9490659f4dccf029b7d/7431cbebb4a74fea0965ac21f2c4de01/f63c0e19559b2cb3622e5a3b481d7cfc,e2bf4b97dbb6b9490659f4dccf029b7d/7431cbebb4a74fea0965ac21f2c4de01/2e9aa20cd62c8f42860812d9e4e6002c
```

Author: linsword13

var/ramble/repos/builtin/modifiers/gcp-metadata/modifier.py

@@ -88,26 +88,34 @@ def gcp_metadata_exec(self, executable_name, executable, app_inst=None):
             )
 
         payloads = [
-            # type, end point, per_node
-            ("instance", "machine-type", False),
-            ("instance", "image", False),
-            ("instance", "hostname", False),
+            # type, end point, per_node, log_name
+            ("instance", "machine-type", False, None),
+            ("instance", "image", False, None),
+            ("instance", "hostname", False, None),
             (
                 "instance",
                 "id",
                 True,
+                None,
             ),  # True since we want the gid of every node
-            ("project", "numeric-project-id", False),
-            ("instance", "attributes/physical_host", True),
+            ("project", "numeric-project-id", False, None),
+            ("instance", "attributes/physical_host", True, None),
         ]
 
-        for type, end_point, per_node in payloads:
+        n_nodes = int(self.expander.expand_var_name("n_nodes"))
+        if n_nodes > 1 and self._usage_mode != "local":
+            # Single-out the vm_id of the executing-node
+            payloads.append(("instance", "id", False, "main-gid"))
+
+        for type, end_point, per_node, log_name in payloads:
             prefix = ""
             suffix = ""
             if per_node:
                 prefix = self.expander.expand_var("{metadata_parallel_prefix}")
                 suffix = self.expander.expand_var("{metadata_parallel_suffix}")
-            log_name = end_point.split("/")[-1]
+            log_name = (
+                log_name if log_name is not None else end_point.split("/")[-1]
+            )
             pre_cmds.append(
                 CommandExecutable(
                     "machine-type",
@@ -221,6 +229,12 @@ def _prepare_analysis(self, workspace):
         group_name="ghostname",
         log_file="{experiment_run_dir}/gcp-metadata.hostname.log",
     )
+    figure_of_merit(
+        "main-gid",
+        fom_regex=r"(?P<gid>.*)",
+        group_name="gid",
+        log_file="{experiment_run_dir}/gcp-metadata.main-gid.log",
+    )
 
     # This returns a list of all known gids in the job
     figure_of_merit(