fix(desktop): add 10s timeout to connection-ipc fetches

[hqhq1025]

Summary

• Wrap all three connection IPC fetch calls (connection:v1:test, models:v1:list, connection:v1:test-active) in a fetchWithTimeout helper backed by AbortController (default 10s).
• Surface AbortError as a clear NETWORK hint so the user sees a timeout message instead of a permanently spinning button.
• Add a unit test that drives the helper with a hanging fake fetch and asserts the abort fires.

Why

Without an AbortSignal, an unreachable or slow baseUrl would pin the renderer's "Test connection" spinner indefinitely (no main-side timeout, no Node-default fetch timeout). 10s is well above the <1s normal /models latency but short enough that the UI fails fast.

PRINCIPLES §5b

• Compatibility green — same response shape, same error codes
• Upgradeability green — CONNECTION_FETCH_TIMEOUT_MS exported as a single source of truth
• No bloat green — pure stdlib, no new deps
• Elegance green — one helper, three call sites

Test plan

• pnpm --filter @open-codesign/desktop exec vitest run src/main/connection-ipc.test.ts — 41/41 pass (incl. new fetchWithTimeout abort test)
• pnpm typecheck — green
• pnpm lint — 0 errors (pre-existing complexity warnings unrelated)

[github-actions]

Findings

• No issues found in added/modified lines with >=80% confidence.

Summary

• Review mode: initial
• No blocker/major/minor/nit issues found in the latest diff.
• Hard-constraint checks in this diff: no new dependencies (license/size impact not introduced), no direct provider SDK imports, no UI token hardcoding touched, and no silent fallback added (errors are surfaced with explicit code/message/hint).
• docs/VISION.md and docs/PRINCIPLES.md: Not found in repo/docs.
• Residual risk: timeout handling is unit-tested via fetchWithTimeout, but there is no integration-style IPC test in this diff proving timeout classification/hints across all three handlers.

Testing

• Not run (automation)

open-codesign Bot