fix: run e2e coverage merge even when a shard flakes (#15342)

<!-- ccr-slack-attribution -->
_Requested by **Christian Byrne** · [Slack
thread](https://comfy-organization.slack.com/archives/C0AP09LKRDZ/p1786872633268499?thread_ts=1786872633.268499&cid=C0AP09LKRDZ)_

## Summary

The e2e coverage merge job only ran when the upstream `CI: Tests E2E`
workflow's overall conclusion was `success`, so a single flaky shard
(16-shard matrix, `fail-fast: false`) skipped merging every other
shard's real coverage — the reported % bounces around a narrow band
instead of climbing as coverage improves.

## Changes

- **What**: `merge` job in
`.github/workflows/ci-tests-e2e-coverage.yaml` now runs on any completed
conclusion except `cancelled` (was: `== 'success'`), so
partial-shard-failure runs still get merged. One-line comment added
explaining why.
- **Breaking**: none

## Review Focus

- The download-artifact step already tolerates a partial set of shard
artifacts (regex-matched artifact names + `if_no_artifact_found: warn`),
and each shard's coverage-upload step runs with `if: always()`, so it
already didn't assume all 16 shards would be present — no change needed
there.
- The "Assert coverage was mapped back to source" step (minimum
source-mapped file count, added in #14950) is untouched — still fails
the job loudly on a degenerate merge.
- `coverage-slack-notify.yaml` already downloads the `e2e-coverage`
artifact with `continue-on-error: true` / `if_no_artifact_found: warn`,
and `scripts/coverage-slack-notify.ts` already treats a
missing/degenerate lcov as "no data" rather than an anomaly — no change
needed there either.
- No new jobs, retries, or timeouts added, so this doesn't add CI
wall-clock time — it's purely a gating/condition change on an existing
job.


---
_Generated by [Claude
Code](https://claude.ai/code/session_01CdcyUD4e6kt6CAuRNkEJ7H)_

Co-authored-by: Claude <noreply@anthropic.com>

Author: claude[bot]

.github/workflows/ci-tests-e2e-coverage.yaml

@@ -15,9 +15,10 @@ permissions:
 
 jobs:
   merge:
+    # Not gated on full success: fail-fast:false means one flaky shard flips the whole run to failure, which would skip merging every other shard's real coverage.
     if: >
       github.repository == 'Comfy-Org/ComfyUI_frontend' &&
-      github.event.workflow_run.conclusion == 'success'
+      github.event.workflow_run.conclusion != 'cancelled'
     runs-on: ubuntu-latest
     timeout-minutes: 10
     outputs: