fix: set a default timeout for REST requests

[EhabY]

Closes #1024

Requests through the CoderApi axios instance had no timeout, so a request hung on a half-open TCP connection (network drop, system sleep) never settled — silently killing every poller that only reschedules after the previous request settles (announcements refresh, workspaces tree poll, OAuth token refresh).

Changes

• defaults.timeout = 60_000 on the shared axios instance in CoderApi.create.
• Unit tests for the default and per-request overrides.

Why no stream exemptions

The issue proposed { timeout: 0 } opt-outs for streaming calls, but they aren't needed: in axios's Node adapter, responseType: "stream" requests are marked done once response headers arrive, so the timeout only bounds the connect/header phase and never aborts an in-flight stream body. SSE streams and CLI downloads just get a bounded connect phase, which is desirable. Build endpoints return as soon as the build is queued, so nothing REST-side legitimately exceeds 60s.

Timeouts surface as regular axios network errors, handled by existing errToStr/retry paths.

Decision log

• No standard to inherit: coder/coder site sets no axios timeout, and Go codersdk uses a bare http.Client{} bounded by per-request contexts.
• 60s over 30s for headroom on slow deployments while still recovering pollers within a minute.
• TCP keepalive on HTTP agents (faster dead-socket detection) deferred as a follow-up.
• A coder.requestTimeout setting rejected as premature configurability.

---

🤖 Generated by Coder Agents on behalf of @EhabY