[0.8] [MOD-18346] Backport CI runner fixes and Debian Trixie

[GuyAv46]

Combined backport of #1032 and #1031 to 0.8. Both fixes are included so CI can validate and merge them together.

• Tag coverage and benchmark EC2 runners at launch with team=ai_and_search and a normalized GitHub actor as owner.
• Run ARM unit tests on GitHub-hosted ubuntu-22.04-arm and remove ARM EC2 provisioning and inherited secrets.
• Replace active Bullseye CI jobs with gcc:14-trixie, rename the Debian installer/workflow, and update workflow dependencies. Preserve release-specific AWS settings and test options.

Fixes MOD-18346.

Validation: parsed workflow YAML, checked local reusable-workflow paths and job dependencies, verified no active Bullseye jobs remain, compiled owner-tag Python, checked EC2 launch tags, and ran shell syntax/whitespace checks. Runtime CI is pending.

---

Note

Low Risk
Changes are limited to CI workflows and test expectations; no production vector-search logic is modified, though CI matrix and EC2 tagging behavior will change on merge.

Overview
CI infrastructure shifts ARM unit tests from ephemeral EC2 runners to GitHub-hosted ubuntu-22.04-arm, dropping start/stop jobs and the need to pass AWS secrets into arm.yml from merge-queue and nightly flows.

Coverage and benchmark EC2 launches now tag instances with team=ai_and_search and a sanitized owner derived from github.actor via a small Python normalization step before machulav/ec2-github-runner starts.

Debian CI replaces active Bullseye jobs with Trixie (gcc:14-trixie): debian13.yml is renamed/retargeted, and merge-queue, nightly, and flow-temp wire trixie instead of bullseye/debian11.

FP16 unit tests in test_override and test_batch_iterator_basic scale vector values by 0.1 to stay under FP16 range limits on ARM/SVE, and override score checks use ASSERT_NEAR with ~0.2% tolerance instead of exact equality.

^{Reviewed by Cursor Bugbot for commit 36d1fb3. Bugbot is set up for automated code reviews on this repo. Configure here.}

Additional merge-queue fix: backport the FP16 test-data scaling from #884 (MOD-13429), already used on newer release branches. This keeps squared L2 distances within FP16 range on the hosted ARM runner; no production distance kernels are changed.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.16%. Comparing base (0178042) to head (36d1fb3).

Additional details and impacted files

@@           Coverage Diff           @@
##              0.8    #1039   +/-   ##
=======================================
  Coverage   97.16%   97.16%           
=======================================
  Files          91       91           
  Lines        4896     4896           
=======================================
  Hits         4757     4757           
  Misses        139      139           

☔ 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.