fix(ci): harden version fetches and auto-open issue on workflow failure

[Hy4ri]

What this does

Same treatment as kimicode-flake#1: the scheduled update workflow dies whenever an external version fetch hits a transient network failure (e.g. TCP connect timeouts to upstream services). This makes the fetch layer resilient and adds automatic failure reporting.

1. Harden all version fetches

• --connect-timeout 10 — fail fast instead of hanging on dropped SYNs
• --retry 3 --retry-delay 3 --retry-all-errors — retry transient errors
• --max-time 30 — cap total time per attempt
• Failure reason captured into a fail_reason job output (via a die() helper) instead of dying silently

Applied to the workflow's check step and the self-fetching update-version.sh paths.

2. Auto-open a failure issue

If the update job fails, a report-failure job opens a GitHub issue with the failure reason + run link:

• Deduped: issues tagged auto-failure — while one is open, later failures comment on it instead of spamming (workflow runs on a schedule)
• Self-healing: when the workflow succeeds again, open auto-failure issues are closed automatically
• The run still goes red (final exit 1 step)
• Adds issues: write permission

Verification

• YAML + bash syntax validated
• Check-step logic simulated locally against real endpoints: all fetches return current versions (no update needed), failure path writes fail_reason + exits 1
• Workflow will be dispatched on this branch to confirm green