fix(dolt): refetch container in setup retry loop

[cofin]

Summary

_provide_dolt_service cached a container handle from _get_container above its retry loop. Under parallel xdist with transient=True, docker_service.run returns success but the container can be auto-removed (Docker remove=True plus a brief Dolt startup blip) between the readiness check passing and the post-run setup exec_run. Every retry then 404s against the same stale ID and the fixture silently yields a broken service, surfacing later as docker.errors.NotFound at test setup.

This recurring flake has hit `tests/test_dolt.py::test_xdist_isolate_server` on the open clientless PRs (e.g. valkey #138 Python 3.11 - 1/3).

Changes

• Refetch the container by name inside the loop so a fresh handle is used each iteration.
• Tolerate `docker.errors.NotFound` and 404/409 `APIError` from `exec_run` so a vanished container re-tries cleanly instead of bubbling up.
• Replace the previous silent-skip-on-missing-container path with an explicit `RuntimeError` after 5 attempts: a genuinely dead container should fail fixture setup loudly instead of producing a half-baked `DoltService`.

Test plan

• `uv run pytest tests/test_dolt.py -v` — all 3 tests pass locally including the previously flaky `test_xdist_isolate_server`.
• `uv run ruff check src/pytest_databases/docker/dolt.py`
• `uv run mypy src/pytest_databases/docker/dolt.py`
• `uv run pyright src/pytest_databases/docker/dolt.py`
• CI green across the full Python matrix.