Build reliability: flaky docker.yml failures on main

[goelakash]

Several docker.yml runs on main fail for reasons unrelated to the change under test. Actions to minimize these flaky build failures:

• Sleep 10s between registry login retries (Sleep between registry login retries #2479)
• Sleep 10s between registry push retries (Sleep 10s between registry push retries #2485). docker push retries back-to-back (~0.25s apart) with no backoff, so a brief quay.io blip fails both attempts. Seen in docker-stacks-foundation tag-push, 2026-06-01 and pytorch-notebook/cuda12 tag-push, scheduled 2026-06-08
• Root-cause the sparklyr/dbplyr [PARSE_SYNTAX_ERROR] ... at or near '"id"' failure in test_spark_r_nbconvert instead of relying on flaky-retry. See all-spark build-test-upload, 2026-06-20
• Investigate the 10activate-conda-env.sh startup stall that fails test_start_container in all-spark (container enters start.sh, runs hooks, never reaches "Executing: jupyter lab"). See scheduled 2026-05-25, scheduled 2026-05-18