[datlechin]

Summary

• tablepro_test_fixtures::start_container takes a closure that builds the container request. When testcontainers reports that pulling the image failed, it builds and starts the request again, up to three times 5 seconds apart.
• Every fixture and the OpenSSH session tests start their containers through it.
• testcontainers pulls once. A Docker Hub pull cut short ("bytes remaining on stream") failed the mongo-7.0 Docker job twice in a few hours: on feat(linux): add the AI assistant #3822 and on main at c6eb271.
• docs/testing.md, Docker or Podman, says so.

Test plan

• New a_pull_that_keeps_failing_is_tried_three_times_and_reported (requires docker): an image tag that does not exist fails with the pull error after three attempts.
• The fixtures' Postgres and OpenSSH tests and the 19 OpenSSH session tests pass through the helper.
• clippy on both crates with -D warnings, cargo fmt --check.