[Feature] Populate [runtimes] in the CI Workbench config so test_runtime_versions runs

Child of #409. Found while measuring the Workbench suite for #539.

workbench/test_runtime_versions.py contributes three tests, and all three skip on every run — not because a Docker Workbench cannot support them, but because the generated config omits one section:

test_r_versions          SKIPPED — No expected R versions specified in vip.toml [runtimes]
test_r_version_in_session SKIPPED — No expected R versions specified in vip.toml [runtimes]
test_python_versions     SKIPPED — No expected Python versions specified in vip.toml [runtimes]

This is free coverage. workbench-smoke.yml writes vip.toml from scratch, so it can populate [runtimes] with whatever the container actually ships.

Scope

RuntimesConfig (src/vip/config.py:390) takes r_versions and python_versions as lists of version strings. The Workbench image installs runtimes under /opt/R and /opt/python, so the workflow can enumerate them from the running container and write them into the config before pytest starts.

Sketch, to be verified against the real image layout:

R_VERSIONS=$(docker exec "$CID" ls /opt/R 2>/dev/null | jq -R . | jq -sc .)
PY_VERSIONS=$(docker exec "$CID" ls /opt/python 2>/dev/null | jq -R . | jq -sc .)

Then emit a [runtimes] block into the generated vip.toml.

Watch out

Populating [runtimes] from the same container the tests then assert against makes the check partly tautological — it proves the versions Workbench reports in-session match what is installed on disk, which is a real check (test_r_version_in_session runs R and compares), but it cannot catch "the wrong runtimes were installed". That is the correct scope for a container smoke test; pinning expected versions is a customer-deployment concern.

These belong in the full tier at minimum. Whether they are fast enough for the PR gate depends on whether each spawns a session — test_r_version_in_session does, so budget roughly 30s for it.

Acceptance

• workbench-smoke.yml writes a [runtimes] block derived from the running container.
• The three test_runtime_versions tests execute rather than skip, verified in the smoke-results.xml artifact.
• The tier placement (gate vs full) is chosen from the measured runtime, and recorded in the workflow.

Read workbench-smoke.yml and RuntimesConfig in src/vip/config.py:390, then inspect workbench/test_runtime_versions.py. Verify the /opt/R and /opt/python layout in the running Workbench container, populate the generated vip.toml [runtimes] block, and choose the tier from measured runtime. Done means all three tests run rather than skip and their results appear in smoke-results.xml.