test: wire up the CT suites ELP reported as unreachable, and bump erlang-ci

[Taure]

Found while auditing whether kura's ELP settings matched reality after widgrensit/asobi#436.

kura's enable-elp-lint: true has been green, but only because the pinned erlang-ci (v2.0.5) installs ELP 2026-02-27, which does not carry the W0008 diagnostic. Under current ELP the tree has 5 lint errors, so a routine pin bump would have turned CI red with no obvious connection to the bump.

Three findings, two different fixes

collect_batches/1 in both coverage suites is a helper that drains batch messages, not a test case. ELP cannot tell: a name/1 in a CT suite that is not in all/0 looks exactly like a test case that never runs. Rather than a suppression comment, it now takes the drain window as a second argument - that says "helper" structurally, and puts the timeout at the call site next to the batch size it is waiting on.

insert_with_opts_returning_rows/1 was the real one, and worse than ELP could tell. It was never in all/0, and had it been, it asserted nothing:

insert_with_opts_returning_rows(_Config) ->
    %% Insert with on_conflict returning empty rows (noop branch, line 175-176)
    ok.

The branch it names is already covered by insert_with_opts_on_conflict_columns two functions above, which inserts the same unique key twice under {{columns, [name, email]}, nothing}. So this is dead weight, not missing coverage, and it is removed rather than wired in - a test that runs and asserts nothing is worse than no test.

Pin bump has to land with it

v2.0.5 -> v2.1.7. That moves ELP to 2026-08-10 (Taure/erlang-ci#71, #72, #73), which is the version that reports W0008 at all - so the bump and this cleanup have to go together or CI goes red on the three above.

v2.1.7 also compiles the test profile before linting, which is what removes the other two errors: kura_custom_type_tests and kura_types_prop_tests both -include_lib("proper/include/proper.hrl"), and under the old rebar3 compile those deps were never fetched, so ELP reported undefined function proper_transformer:parse_transform/2 for both. Measured on this tree: 5 errors before, 3 after the profile change, 0 after this cleanup.

Verification

• kura_coverage_SUITE + kura_coverage2_SUITE: 87/87.
• elp lint: 0 errors (29 warnings, unchanged), was 5.
• rebar3 fmt --check, xref: clean.
• kura_resilience_tests fails locally (2 failures, 1 cancelled) because it takes docker-compose control and another checkout holds port 5555. It fails identically on unmodified main and is skipped in CI, so it is environmental and unrelated.

Not in this PR

enable-elp-eqwalize stays false: 92 errors on this tree (76 incompatible_types, 16 redundant_nowarn_function). Worth noting the 16 redundant_nowarn_function are cheap - eqwalizer saying suppressions are no longer needed - so the real work is closer to 76. asobi tracks the equivalent in widgrensit/asobi#435 if you want the same shape of plan here.

[github-actions]

🟢 Code Coverage — 92.4%

2612 of 2826 lines covered.

---

🟡 ELP Lint — 29 warnings

34 diagnostics found. See job logs for details.