test: flaky 'analyze deps: repository size detection' — commit exit 128 in CI loop

[elifarley]

Flaky test

tests/unit/test_analyze_deps_performance.bats — "hug analyze deps: repository size detection" fails intermittently in CI with:

not ok 596 hug analyze deps: repository size detection in 588ms
# (in test file tests/unit/test_analyze_deps_performance.bats, line 102)
#   `git commit -m "commit $i"' failed with status 128

The test runs a 50-iteration commit loop (for i in {1..50} then stages + commits per iteration). CI logs show commits 1-11 succeeding before the failure — the git commit invocation exits 128 mid-loop, which is a resource/timing pressure signature (not a logic error in the code under test).

Evidence it is flaky (not a real regression)

Surfaced while addressing review feedback on #178, which touches only git-wtdel/git-wtc (no causal path to git analyze-deps):

• commit e35f32f: FAILED test (unit)
• commit d9a873e: PASSED test (unit)
• commit 4f93ac6 (the Worktree family hardening: fix P0 repo deletion, unify batch/exit-code/JSON contracts #178 review fixes): FAILED on first run, PASSED on gh run rerun --failed — no code change between the two outcomes.
• The test passes 3/3 locally.

Suggested fix direction

• Reduce or parallelize the commit-loop pressure, or wrap each commit in a retry.
• Investigate whether the run_with_timeout wrapper can cascade a timeout into a child commit exiting 128.
• Consider seeding git identity explicitly in the test's setup() to rule out identity-related 128s (the same root cause category fixed by BATS test setup leaks scratch files (newfile.txt, staged.txt) into the worktree — create_test_repo output not captured #180/test(dispatcher): isolate per-test CWD via setup()/teardown() #185 for the dispatcher suite).

Motivated by

#178 — the spurious failure blocked merge gating until a manual gh run rerun cleared it.