fix(aimlapi): stop issue floods and remove sunset models

[rekram1-node]

Summary

Corrects the issue flood introduced by #9531. The runner includes every skipped remote ID in the missing-model issue batch when skipCreates: true, independently of the provider's missingModelID filter. The earlier tests checked the provider callback but not actual issue admission through the runner.

• Disable automatic missing-model issues for AI/ML API (trackMissingModels: false) until its backlog is audited. Existing updates/removals continue; omitted models are report-only. Other providers are unaffected.
• Add a regression test through the actual syncProvider runner with openIssues: true, covering both untranslated IDs and translated new cards blocked by skipCreates. Neither is admitted to issues.
• Read the public deprecation feed alongside the catalog, matching canonical IDs and aliases. Exclude withdrawn models and deprecated models past their shutdown date; retain superseded aliases that still resolve. Mark future/undated deprecated models as deprecated.
• Remove gpt-5.1-codex, gpt-5.1-codex-mini, and gpt-5.2-codex: the host feed lists their shutdown date as 2026-07-23, despite retaining catalog/docs entries.
• Fail the sync before writing if the deprecation feed cannot be fetched or validated.

Cleanup already performed

Closed 114 AI/ML API missing-model issues and 75 automatically generated follow-up PRs from the batch. No unrelated provider issues or PRs were closed. Checked that none remained open and no issue-fixer jobs were active at inspection.

Checks

• bun validate passes.
• bun test test/aimlapi.test.ts from packages/core: 20 pass.
• Live bun models:sync aimlapi --dry-run --open-issues: zero issue admissions, zero creates/updates, and exactly the three sunset removals now included in this PR.

Public catalog presence and a docs endpoint returning 200 do not establish inference availability. This PR uses the host's explicit lifecycle records rather than guessing retirement from model age.

[github-actions]

No actionable findings.