Stabilize AWQ runtime loading and reproducible image build baseline

[parkcheolhong]

PR Title

Stabilize AWQ runtime loading and reproducible image build baseline

Summary

This PR packages the AWQ recovery work into a reviewable unit and includes reproducible runtime/image updates plus operational evidence documents.

Why

• AWQ load path previously failed intermittently under generic model load flow.
• Container builds were unstable due to package metadata/build path issues.
• Operational closure needed explicit restart + health evidence and traceable release notes.

Scope Of Changes

• Runtime
  • Added AWQ-first model loading path with safe fallback behavior.
• Image/Dependency
  • Hardened dependency installation for gptqmodel and metadata-broken packages.
  • Added pypcre patch helper flow for deterministic builds.
• Ops Documentation
  • Synchronized AWQ recovery checklist with post-restart verification evidence.
  • Added commit plan and release notes for traceability.
• Repository Baseline
  • Added compose profiles, env presets, operation docs, monitoring/nginx configs, and lightweight web UI assets.

Commit Units

• 5eb1b53 fix(runtime): add AWQ dedicated loader path and keep safe fallback
• 896c7c1 fix(image): harden dependency install path for gptqmodel and broken package metadata
• 62579fe docs(ops): sync AWQ recovery checklist with post-restart validation evidence
• 3b35d42 chore(repo): add baseline configs, compose profiles, and operation docs
• 30d3bac feat(observability): add monitoring stack and nginx gateway configs
• 785a734 feat(web-ui): add lightweight model server dashboard
• c314966 docs(release): add 2026-04-27 AWQ recovery release notes

Validation Evidence Checklist

• AWQ loader success log confirmed in runtime startup flow.
• Health endpoint success observed multiple times after startup stabilization.
• Post-restart health verification reflected in ops checklist docs.
• Tagged milestone commits created and pushed.

Tags Included

• awq-runtime-loader-v1 -> 5eb1b53
• awq-image-hardening-v1 -> 896c7c1
• awq-ops-evidence-v1 -> 62579fe

Risk And Rollback

• Risk: Runtime behavior changes in model load path can impact non-AWQ model variants.
• Mitigation: Fallback path retained; deploy on branch and validate health before promote.
• Rollback: Revert from tag boundaries or rollback to pre-5eb1b53 state.

Reviewer Focus

• AWQ-first loader path and fallback branching correctness.
• Dockerfile dependency hardening correctness/reproducibility.
• Operational evidence consistency between checklist and release notes.

[sourcery-ai]

Sorry @parkcheolhong, your pull request is larger than the review limit of 150000 diff characters

[parkcheolhong]

모두 검토했습니다

[parkcheolhong]

모두 확인하였습니다, 잘못된 것이 있으면 수정해주세요.

[parkcheolhong]

2026/04/27일자로 ui/ux 브라우저 변경했습니다.

[parkcheolhong]

확인했습니다.