fix(consume): harden enginex hive reporting against connection errors

[danceratopz]

WIP: please don't review yet; pending self-review

Related: ethereum/hive-python-api#18

Description

Complementary simulator-side hardening for the TIME_WAIT ephemeral-port exhaustion behind the sporadic Nethermind consume-enginex dashboard failures on 2026-07-18/19 (4–5 spurious "Test was terminated by host" fails and 85/40 silently missing test results per run). The root fix pools the Hive API connections in ethereum-hive (hive-python-api#18); this PR makes the enginex reporting paths resilient and makes any remaining silent test loss loud:

• Retry with backoff on connection errors in the per-test reporting paths: start_test/end_test in the hive_test fixture and register_multi_test_client in the enginex conftest (5 attempts, 0.5 s initial backoff, ~7.5 s total). Connection errors are raised before the request is sent, so retrying cannot duplicate a call. Works against the released ethereum-hive==0.1.0a5, i.e. needs no package release; with the package fix it is belt-and-suspenders.
• Collected-vs-reported assertion (enginex only, via config.assert_reported_test_count): each xdist worker counts the test results it successfully reports to hive (plus setup-skipped tests); the last worker to finish aggregates all workers' counts and compares against session.testscollected before ending the suite. On a shortfall it (a) reports a failing reported-test-count-check meta-test case to hive, making the loss visible in the hive UI where lost tests are otherwise silently missing, and (b) raises HiveReportedTestCountError, erroring the pytest session. Skipped on interrupted runs (-x/--maxfail/^C); only a shortfall fires, so skip edge cases cannot false-fail. This also surfaces the known singleton-group client-startup flake instead of letting it silently shrink the listed test count.

Helpers live in plugins/pytest_hive/reporting.py (a non-plugin module so the enginex conftest can import it without disabling assert rewriting for the plugin), with unit tests covering the loss path a healthy run cannot exercise.

Validated with hive --dev + nethermind (12,888 Cancun tests, -n 4): verdicts identical to baseline, check reports 12,888/12,888, peak TIME_WAIT toward the hive API 9,684 → 66 with the patched package.

Follow-up (separate PR once ethereum-hive==0.1.0a6 is on PyPI): bump the pin in packages/testing/pyproject.toml to >=0.1.0a6,<1.0.0.

Related Issues or PRs

Root fix: hive-python-api#18 (root-cause diagnosis: nethermind enginex fails 2026-07-18/19 — simulator ephemeral-port exhaustion, not a client bug).

Checklist

• Ran fast static checks to avoid CI fails, see Code Standards & Verifying Changes: just static
• PR title has the form <type>(<area>): <title>, where <type> and <area> come from an appropriate C-<type>, respectively A-<area>, label. The title should match the target squash commit message.

Cute Animal Picture

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.50%. Comparing base (f1aa970) to head (f8f5593).
⚠️ Report is 8 commits behind head on forks/amsterdam.

Additional details and impacted files

@@               Coverage Diff                @@
##           forks/amsterdam    #3335   +/-   ##
================================================
  Coverage            93.50%   93.50%           
================================================
  Files                  624      624           
  Lines                37070    37070           
  Branches              3394     3394           
================================================
  Hits                 34661    34661           
  Misses                1653     1653           
  Partials               756      756           

Flag	Coverage Δ
unittests	93.50% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.