feat: separate benchmark for profiling the stdlib per-file

tests/bench/accumulate_profile.py

@@ -10,4 +10,7 @@ for m in re.findall("^([^\n\d]+)([\d.]+)(m?)s$", data, re.MULTILINE):
     cats[m[0].strip()] += float(m[1]) * (1e-3 if m[2] else 1)
 
 for cat in sorted(cats.keys()):
-    print(f"{cat!r}: {cats[cat]}")
+    cat2 = cat
+    if len(sys.argv) > 1:
+        cat2 = f"{sys.argv[1]} {cat}"
+    print(f"{cat2!r}: {cats[cat]:f}")

tests/bench/full-stdlib.exec.yaml

@@ -0,0 +1,17 @@
+- attributes:
+    description: stdlib
+    tags: [slow]
+    time: &time
+      #runner: time
+      # alternative config: use `perf stat` for extended properties
+      runner: perf_stat
+      perf_stat:
+        properties: ['wall-clock', 'task-clock', 'instructions']
+      rusage_properties: ['maxrss']
+  run_config:
+    <<: *time
+    cwd: ../../src/
+    cmd: |
+      find . -name '*.lean' -exec bash -c 'lean --stats -Dprofiler=true -Dprofiler.threshold=9999 {} 2>&1 | ../tests/bench/accumulate_profile.py {}' \;
+    max_runs: 2
+    parse_output: true