fix(ros): make the interop gate count interop tests

[YuanYuYuan]

Summary

The interop gate merged in #271 does not count interop tests. It asserts a non-zero package total and then reports it as interop:

Summary [47.809s] 125 tests run: 125 passed, 3 skipped
125 ROS interop tests ran against rmw_zenoh_cpp.

Only 41 of those are interop binaries — pubsub_interop 14, demo_nodes 9, service_interop 7, type_description_interop 4, action_interop 4, parameter_interop 3. The other 84 are lifecycle, cache, parameter_tests and friends. Delete every interop test and the gate still prints a confident number.

That is the defect #271 was written to remove, reintroduced in the fix.

Key Changes

• Count the interop binaries by name; require each to be present. A run of 84 passing non-interop tests now fails, instead of reporting "84 ROS interop tests ran". type_description_interop is exempt on Humble, where the suite is #![cfg(not(ros_humble))] because there is no type description service to interoperate with; every other suite is mandatory on every distro.
• Fail when a test self-skipped for a missing ros2 CLI. Each interop test returns early — still passing, still counted — when check_ros2_available() is false. test(ros): fail the interop job when it runs no tests #271 named this failure mode in its own description and did not close it. check_ros2_available uses Command::new("ros2").output().is_ok(), which is true if the process merely spawned, so a runner with a broken rmw_zenoh_cpp produces a green job full of skips.
• Gate the run that decided the outcome. The retry path re-ran nextest but the gate parsed the first attempt: a passing retry was validated against discarded output, and a first attempt that died before producing a summary failed an otherwise-green run.
• Drop the 0 tests run branch. cargo-nextest 0.9.138 defaults --no-tests to fail, so it was unreachable. The repo already knew this — .github/workflows/test.yml passes --no-tests=warn precisely because the default fails.

Output becomes a per-suite breakdown rather than one aggregate.

What fails without this

On main today, the gate's own output is the evidence: it printed 125 ROS interop tests ran against rmw_zenoh_cpp for a run containing 41.

The gate then detected a real empty suite on its own first CI run. The humble job failed with:

ROS interop suites produced no tests: type_description_interop.
A pass here would be vacuous.

That was a correct detection of a genuinely-empty test binary, caught by the new check and invisible to the old one — which had reported the same humble run as 116 ROS interop tests ran. It turned out to be a legitimate distro difference rather than a defect, and is now exempted, but it is the proof the detector detects rather than merely never firing.

Verified in five directions against real nextest output:

input	result
healthy jazzy run, all six suites present	pass — prints the 41-test breakdown
healthy humble run, five suites	pass
84 passing tests, no interop suites	fail — names the missing suites
healthy output plus one ros2 CLI not available line	fail
empty output	fail — no summary

Notes

Found by an adversarial review of the merged PR. The remaining gap, not addressed here: the four Interop tests with ROS 2 * jobs in test.yml run nextest directly and are ungated — one of them produced Summary [0.000s] 0 tests run: 0 passed, 13 skipped and passed, on the very commit that merged #271.

Breaking Changes

None. CI-only; the gate becomes stricter.