ci: balance the backend test shards by measured time, not test count

Backend Tests (shard 1/4) timed out after 10 minutes on e8b901f54
("Updated BACKERS"), a docs-only commit. The step was not hung: it
reached 98%, every test passing, and was killed about 12 seconds short
of finishing.

pytest-split balances by duration only when it has a durations file,
and there was none. Without one it splits by test COUNT -- 2926 / 2926
/ 2926 / 2925, exactly 25% each, which is what the comment here claimed
was good enough. Count is not time. Measured over the full suite
(11703 tests, 1045.6s):

    shard 1   2926 tests   658.2s
    shard 2   2926 tests   173.1s
    shard 3   2926 tests   113.3s
    shard 4   2925 tests   101.1s

Shard 1 was carrying 63% of the suite's runtime -- a 6.5x spread -- and
had been walking toward the cap for weeks: 475s, 404s, 445s, then 587s
on 08-30, thirteen seconds under it, and 613s here. CI printed the
reason in its own log on every run: "[pytest-split] No test durations
found."

Commit tests/.test_durations and pass --durations-path explicitly:

    shard 1   1273 tests   261.6s
    shard 2   1070 tests   261.5s
    shard 3   1522 tests   262.0s
    shard 4   7838 tests   260.6s

A 1.01x spread. The lopsided test counts are the point: shard 4 takes
thousands of fast unit tests, shard 1 keeps the slow integration ones.
The four groups still partition the suite exactly -- union is 11703
with nothing dropped or duplicated, and all four pass.

--durations-path has to be explicit because pytest-split defaults it to
$CWD/.test_durations, and the two matrices run from different
directories: the native job from backend/, the Docker job from the
image root. Left implicit, the Docker one finds nothing and silently
falls back to the count split. Node IDs match across both because
backend/tests/pytest.ini pins rootdir to backend/tests either way, so a
single file serves them both; verified the file clears .dockerignore
and lands at /app/backend/tests/.test_durations at full size.

timeout-minutes 10 -> 15 for headroom, since the file goes stale as
tests are added. Staleness degrades slowly rather than breaking --
unknown tests are treated as average.

.github/workflows/ci.yml

@@ -124,21 +124,35 @@ jobs:
           pip install -r requirements-dev.txt
 
       - name: Run tests (shard ${{ matrix.shard }}/4)
-        timeout-minutes: 10
+        timeout-minutes: 15
         run: |
           cd backend
           # -v dropped: 5300+ "PASSED foo::bar" lines per worker eat 30-60s
           # of stdout I/O time on 2-vCPU runners. --tb=short is enough.
-          # --splits 4 --group N uses pytest-split to slice the collected
-          # test set roughly evenly across the 4 matrix shards; first run
-          # is name-hash-based, subsequent runs improve via .test_durations
-          # if you ever commit one (we don't — even the naive hash split
-          # gets us ≈25% per shard given the test mix here).
+          #
+          # --splits 4 --group N slices the suite across the 4 matrix shards.
+          # Without a durations file pytest-split balances by test COUNT, which
+          # is not the same thing as time: measured 2026-08-31, that put 658s of
+          # the suite's 1046s into shard 1 and 101s into shard 4 -- a 6.5x spread
+          # that walked shard 1 into the old 10-minute timeout on a docs-only
+          # commit. tests/.test_durations is committed so the split is balanced
+          # by measured time instead (1.01x spread). Shard test counts are
+          # lopsided by design once it is used: shard 4 takes thousands of fast
+          # unit tests, shard 1 keeps the slow integration ones.
+          #
+          # --durations-path must be explicit: pytest-split defaults it to
+          # $CWD/.test_durations, and this job runs from backend/ while the
+          # Docker job below runs from the image root. Left implicit, one of
+          # them silently finds no file and falls back to the count split.
+          # Regenerate after adding slow tests:
+          #   cd backend && python -m pytest tests/ --store-durations \
+          #     --durations-path tests/.test_durations
           python -m pytest tests/ \
             --tb=short \
             --timeout=60 --timeout-method=thread \
             -n auto \
-            --splits 4 --group ${{ matrix.shard }}
+            --splits 4 --group ${{ matrix.shard }} \
+            --durations-path tests/.test_durations
 
   # ============================================================================
   # Frontend Checks
@@ -365,7 +379,8 @@ jobs:
               --timeout=60 --timeout-method=thread \
               -p no:cacheprovider \
               -n auto \
-              --splits 4 --group ${{ matrix.shard }}
+              --splits 4 --group ${{ matrix.shard }} \
+              --durations-path backend/tests/.test_durations
 
   docker-test:
     name: Docker Build

.pre-commit-config.yaml

@@ -31,7 +31,17 @@ repos:
         args: ['--maxkb=1000']
         # CHANGELOG.md is intentionally large (detailed per-release entries over
         # many versions); exempt it while keeping the 1 MB guard for everything else.
-        exclude: ^(static/assets/|CHANGELOG\.md$)
+        #
+        # backend/tests/.test_durations is ~1.5 MB and cannot be made smaller:
+        # its 11703 pytest node IDs are 1160 kB of keys before a single duration
+        # value is written, so no encoding fits under the guard (compact JSON
+        # with 3-decimal values still measures 1263 kB). Pruning fast tests is
+        # worse than useless -- pytest-split gives a test missing from the file
+        # the AVERAGE of the entries that remain, so dropped 1 ms tests come back
+        # weighted at ~89 ms and the balance it exists to provide degrades.
+        # It compresses to ~314 kB in the object store. Regenerate with
+        # --store-durations; see the backend test job in .github/workflows/ci.yml.
+        exclude: ^(static/assets/|CHANGELOG\.md$|backend/tests/\.test_durations$)
       - id: check-merge-conflict
       - id: debug-statements
       - id: detect-private-key