Cut the mutation testing timeout from 60s to 20s

.cargo/mutants.toml

@@ -31,11 +31,18 @@ examine_globs = [
 exclude_globs = ["src/testutil.rs"]
 
 # The unit tests finish in well under a second, so anything still running after
-# a minute is a mutant that made the code loop rather than one under test. There
+# twenty is a mutant that made the code loop rather than one under test. There
 # are twenty of those in the parser, which is what a mutated `pos += 1` in a
 # byte scanner does, and they are reported as timeouts rather than survivors.
+#
+# Sixty seconds here was the first guess and it was too generous to be free.
+# Twenty timeouts at a minute each is twenty minutes of a run that should take
+# twelve, and they do not spread evenly across the shards: the run on 2026-08-22
+# had five land consecutively in shard 2, which took the job out. Twenty seconds
+# is still more than thirty times what the suite needs, and it costs the whole
+# set of them under seven minutes.
 timeout_multiplier = 5.0
-minimum_test_timeout = 60
+minimum_test_timeout = 20
 
 # Ten mutants survive in this scope as of the run that wrote this file, out of
 # roughly 250. They fall into three groups, named rather than chased:

.github/labeler.yml

@@ -42,6 +42,10 @@ tests:
           - "fuzz/**"
           - "scripts/crossvalidate.R"
           - "src/testutil.rs"
+          # cargo-mutants' scope and timeouts. Left out at first, so the pull
+          # request that tuned them arrived with no label at all, which is the
+          # failure this file's own comment warns about one section down.
+          - ".cargo/**"
 
 packaging:
   - changed-files: