Support graphing multiple run comparisons

When graphing metrics from two runs, the timestamps rarely align; so we add a
`relative` option to convert the absolute metric timestamps into relative
delta seconds from each run's start.

backend/app/services/crucible_svc.py

@@ -64,14 +64,21 @@ class GraphList(BaseModel):
     omitted if all Graph objects specify a run ID. (This is most useful to
     select a set of graphs all for a single run ID.)
 
+    Normally the X axis will be the actual sample timestamp values; if you
+    specify relative=True, the X axis will be the duration from the first
+    timestamp of the metric series, in seconds. This allows graphs of similar
+    runs started at different times to be overlaid.
+
     Fields:
         run: Specify the (default) run ID
         name: Specify a name for the set of graphs
+        relative: True for relative timescale in seconds
         graphs: a list of Graph objects
     """
 
     run: Optional[str] = None
     name: str
+    relative: bool = False
     graphs: list[Graph]
 
 
@@ -1208,6 +1215,7 @@ async def get_runs(
                         "params": iparams,
                     }
                 )
+            run["iterations"].sort(key=lambda i: i["iteration"])
             run["params"] = common.render()
             try:
                 run["begin_date"] = self._format_timestamp(run["begin"])
@@ -1904,10 +1912,19 @@ async def get_metrics_graph(self, graphdata: GraphList) -> dict[str, Any]:
             x = []
             y = []
 
+            first = None
+
             for p in sorted(points, key=lambda a: a.begin):
-                x.extend(
-                    [self._format_timestamp(p.begin), self._format_timestamp(p.end)]
-                )
+                if graphdata.relative:
+                    if not first:
+                        first = p.begin
+                    s = (p.begin - first) / 1000.0
+                    e = (p.end - first) / 1000.0
+                    x.extend([s, e])
+                else:
+                    x.extend(
+                        [self._format_timestamp(p.begin), self._format_timestamp(p.end)]
+                    )
                 y.extend([p.value, p.value])
                 y_max = max(y_max, p.value)