fix: add 30s timeout to GoldSky subscriber fetch

[realfishsam]

Fixes #512

[realfishsam]

PR Review: PASS (NOT VERIFIED)

What This Does

Adds a 30-second timeout to GoldSky subscriber GraphQL fetch calls using AbortSignal.timeout(), combined with the existing caller-provided abort signal via a shared AbortController. Prevents indefinite hangs when the GoldSky subgraph endpoint is unresponsive.

Blast Radius

• Core: core/src/subscriber/external/goldsky.ts only
• No OpenAPI, SDK, or type changes

Findings

1. Signal composition is correct: The pattern of creating a combinedController and wiring both the caller's signal and AbortSignal.timeout(30_000) to abort it is the standard approach. The finally block properly removes both event listeners to avoid leaks.
2. Node.js compatibility: AbortSignal.timeout() requires Node 17.3+. The project requires Node >=18.0.0, so this is safe. This is the first use of AbortSignal.timeout() in the codebase.
3. TimeoutError handling: The catch block now also suppresses TimeoutError (in addition to AbortError), returning null silently. This is consistent with the existing behavior of treating fetch failures as non-fatal for polling subscribers. However, a logger.debug for timeouts specifically might help debugging slow endpoints in production. Non-blocking.
4. Hardcoded 30s timeout: The timeout is not configurable. This matches the existing DEFAULT_WATCH_TIMEOUT_MS constant used elsewhere, but a config option would be more flexible. Non-blocking.

PMXT Pipeline Check

• Field propagation: N/A
• OpenAPI sync: N/A
• Type safety: OK

Semver Impact

patch -- internal resilience fix, no API or behavioral change

Risk

Not verified against a live GoldSky endpoint. The 30s timeout is reasonable but may be too aggressive if the subgraph legitimately takes longer under load. Monitor after deploy.