ci(workflows): unblock LLM PR trigger by trimming comments in oversized run block

[gianni-cor]

🎯 What problem does this PR solve?

• On PR Trigger (LLM) is currently broken on main for every PR that touches the LLM package. Example failure: run 25332266722.
• After Create new buckets to run tests in independent processes #1889 was merged, the Create and Upload Test Spec step's run: | block in .github/workflows/integration-mobile-test-qvac-lib-infer-llamacpp-llm.yml reached 21,074 chars (YAML-parsed) — 74 over GitHub Actions' 21,000-char single-expression limit.
• Because the LLM trigger uses pull_request_target + workflow_call into that file (loaded from main), the parser fails fleet-wide with:

  (Line: 914, Col: 14): Exceeded max expression length 21000
  

  No jobs run; the workflow exits with failure and 0 billable seconds.
• Currently blocked PRs include Sync with latest fabric-8189 version on LLM/Embed/NMT  #1874 (Qwen3.5 / Gemma4 / PaddleOCR-VL tests) and several others.

📝 How does it solve it?

• Removes 32 in-block comment lines from the offending step's run: | block — pure prose, no logic changes.
• Comment groups removed (all inside the Create and Upload Test Spec step):
  • 4 lines narrating per-platform branching
  • 7 lines describing Android crash detection / placeholder substitution
  • 1 line about iOS host (macos_sequoia) requirement
  • 15 lines describing iOS crash flow / WDA / timeout
  • 5 lines describing per-split make_split injection
• Brings the YAML-parsed run: payload from 21,074 → 19,008 chars (1,992 under the 21k limit).
• Diff: +0 / -32.

🧪 How was it tested?

• Re-parsed the post-edit workflow file and confirmed the largest run: block is now 19,008 chars vs the 21,000 limit.
• Verified diff against upstream/main is comments-only (git diff --stat: 32 deletions, 0 insertions).
• Behavior the comments described (5 s setTimeout exit, flushBareLog Appium pull_file, __TEST_FILTER__ / __QVAC_PERF_* placeholder substitution, etc.) is unchanged — the step still runs the same shell, generates the same testspec YAMLs, and uploads to the same Device Farm project.
• Real validation will come from On PR Trigger (LLM) actually starting jobs on this PR (which is the failing path on main).

🔭 Follow-up (not in this PR)

• The run: block has been creeping toward the 21k limit for a while: 19,410 (4/4 - Remove unnecessary NPM tokens #1591) → 21,103 (fix(ci): pull bare_console.log from iOS app sandbox on crash #1804, first time over) → 22,734 (QVAC-18111 infra[notask]: bridge QVAC_PERF_RUNS to mobile test app via pushFile #1840) → 23,435 (Create new buckets to run tests in independent processes #1889, full break). Comment-trim buys headroom but not much.
• Suggested next step: extract the Android + iOS WDIO_CONFIG JS blobs (and ideally the long printf testspec template) into files under .github/scripts/ or packages/qvac-lib-infer-llamacpp-llm/test/mobile/ and cat / source them in the workflow. That permanently caps run-block size and lets the documentation comments come back in a place where they don't count against the 21k limit.

[gianni-cor]

/review

[gianni-cor]

/review

[github-actions]

Tier-based Approval Status

**PR Tier:** TIER1

**Current Status:** ✅ APPROVED

**Requirements:**
- 1 Team Member approval ❌ (0/1)
- 1 Team Lead OR Management approval ✅ (2/1)

**Bypass rule:** Triggered (2+ Team Lead approvals (Tier 1 exception)). This PR is approved regardless of tier.

---
*This comment is automatically updated when reviews change.*

[gianni-cor]

/review