fix(extensions-portal): increase install timeout and nginx proxy_read_timeout

[yasinBursali]

What

Increase the frontend mutation timeout from 15 s to 120 s and add a descriptive "Installing…" hint during extension installs. Also raises the nginx proxy_read_timeout from the 60 s default to 180 s so the reverse proxy never cuts the connection before the frontend does.

Why

The frontend AbortSignal.timeout(15000) fired before the backend could complete a Docker image pull. The install would succeed in the background but users saw a false "Failed to install" error. The nginx default of 60 s created a secondary bottleneck in production that the frontend timeout fix alone would not have resolved.

How

• Extensions.jsx: AbortSignal.timeout(15000) → AbortSignal.timeout(120000)
• Extensions.jsx: added mutatingAction state; Install button shows "Installing…" while in flight
• nginx.conf: added proxy_read_timeout 180s to the /api/ location block

Scope

All changes are within dream-server/extensions/services/dashboard/.

Testing

• Install an extension with an uncached Docker image (e.g. AnythingLLM) — verify no false error at 15 s, "Installing…" label appears
• Enable/disable an extension — verify spinner only (no "Installing…" label)
• Test in the containerised dashboard (not just vite dev) to exercise the nginx timeout

Review

Critique Guardian: APPROVED (after nginx companion fix added)

[Lightheartdevs]

Audit review: Incremental fix building on #673. Changes look correct. Must merge in sequence after #673. LGTM.

[Lightheartdevs]

Approved. Needs rebase onto main to fix tier-*-env-validation failures (missing recent schema changes). Merges in sequence after #673.

[Lightheartdevs]

Hey Yasin — good progress, here's where everything stands:

Merged today: #635, #669, #670, #671, #672, #673, #706, #708, #709, #710

Need rebase (merge conflicts): #711, #712, #713, #721 — these conflict after the earlier chain merged. Once rebased they're all approved and ready to go.

Still need fixes from earlier reviews:

• fix: fail-closed on missing WEBUI_SECRET and add open-webui memory limits #698 — CI fix needed: add WEBUI_SECRET=ci-placeholder to the validate-compose.yml workflow so docker compose config doesn't choke on the :? syntax
• fix: add authentication to GPU detail, topology, and history endpoints #696 — needs a companion frontend change to pass auth headers to the GPU endpoints, otherwise the dashboard GPU panel breaks with 401s after merge
• fix: use empty secret_key placeholder for SearXNG fail-closed behavior #703 — sent back: empty secret_key is a downgrade from the placeholder (SearXNG silently accepts it with zero entropy). Consider the :? compose env var approach instead.
• fix: replace qdrant healthcheck with perl io::socket instead of bash /dev/tcp #704 — sent back: perl likely not in the qdrant image, and the endpoint should be / not /healthz

Almost there — 4 rebases and you're done with the extensions portal chain.