why: test_freeze_config and test_freeze_logs_debug each slept a
fixed time.sleep(0.50) to let tmux settle after build() before
freeze() — ~1.0s of dead wait per run.

what:
- Replace both sleeps with retry_until polling on
  len(session.windows) >= len(session_config["windows"])
- Import retry_until (libtmux), drop now-unused import time
- Suite 39.69s -> 38.53s (-1.16s), above the 0.68s k*MAD noise band

why: the suite is ~88% call-bound across independent, tmux-driven
test bodies (each test spawns its own daemon on a unique socket),
so it parallelizes cleanly. Measured: serial 39.69s -> ~16-19s
under capped xdist (best 13.7s) ~= 2.1-2.9x, far above the 0.68s
noise band; fully green under -n 8 in clean CI conditions.

what:
- Add pytest-xdist to the dev and testing dependency groups
- Add a `just test-parallel` recipe with a capped worker count
  (-n 8); `just test` stays serial
- Keep -n out of addopts: -n auto (=cores) thrashes at full core
  saturation (~40s) and -n breaks --pdb/-x ergonomics

why: CLI color is decided live from FORCE_COLOR/NO_COLOR (see
tmuxp._internal.colors.Colors). A contributor who exports
FORCE_COLOR in their shell gets ANSI-wrapped help text, so
test_help_examples.py fails (len(examples) == 0). The same leak
made the suite look order-dependent under xdist when FORCE_COLOR
reached the workers.

what:
- Add autouse _pin_test_color_env clearing FORCE_COLOR/NO_COLOR for
  every test; color-specific tests still set them explicitly and
  monkeypatch restores the contributor's values at teardown
- Suite now green with FORCE_COLOR exported and under xdist
  regardless of ambient color env

why: CLAUDE.md mandates functional tests only ("avoid class TestFoo
groupings"). test_finders_local.py had two class-based groupings,
the only such violation in the suite.

what:
- Flatten TestFindLocalWorkspaceEdgeCases (7 methods) and
  TestLocalWorkspaceFilesConstant (2 methods) into module-level
  def test_* functions; drop the self receiver
- Rename the constant tests to test_local_workspace_files_constant_*
  so they read at module scope without the class for context
- No behavior change: same tests collected, assertions verbatim

why: without --strict-markers an unregistered or mistyped marker
(e.g. @pytest.mark.flakey) is a silent warning, so a typo'd mark
quietly stops applying. Strict mode turns it into a hard collection
error.

what:
- Add --strict-markers to [tool.pytest.ini_options] addopts
- No markers ini entry needed: the only custom marker (flaky) is
  registered by pytest-rerunfailures; there are no project-defined
  custom markers
- Verified: suite green and an unregistered mark now errors at
  collection

why: WorkspaceBuilder.build() already calls _wait_for_pane_ready()
for each normal pane, so the session is fully materialized when
build() returns. The post-build retry_until on window count was a
no-op (windows register synchronously) and its comment overstated
what it did.

what:
- Remove the retry_until window-count waits from test_freeze_config
  and test_freeze_logs_debug, plus the now-unused retry_until import
- build() flows straight to freeze(); the existing freezer tests
  remain the regression coverage

why: test_at_filesystem_root requested tmp_path and monkeypatch but
used neither — it exercises traversal against the real filesystem
root and patches nothing.

what:
- Remove the unused tmp_path/monkeypatch parameters from the test
  signature

why: the comment said "should find both" above an assertion that
tolerates more than two results — when traversal continues past
home it can also pick up configs in ancestor directories.

what:
- Reword the comment to match the >= 2 assertion

why: test_local_workspace_files_constant_is_list only asserted the
constant is a list of length 3 — both already proven by the sibling
test_local_workspace_files_constant_order via its equality assertion
against the list literal.

what:
- Remove the subsumed is_list test