Support per-operation API request timeouts through Forge metadata #26

Problem

cf workers versions profile supports CPU and heap captures lasting up to 50 seconds, but cf currently applies a 30-second default API request deadline. A valid capture can therefore be aborted before the profiling operation finishes.

cf workers versions profile latest \
  --worker-id "$WORKER_ID_OR_NAME" \
  --duration-ms 50000 \
  --profile-type cpu > worker-cpu.pprof.gz

See the profiling API documentation and announcement.

Workaround in cloudflare/cf#257

cloudflare/cf#257 raises the shared default API timeout from 30 to 60 seconds. This gives a maximum-length profiling capture 10 seconds of additional request budget and applies consistently to SDK and raw-response requests.

This is a general default increase rather than an endpoint-specific timeout policy. It addresses the immediate mismatch, but other operations may need different timeouts.

What we need from Forge

Something like:

x-forge-request-timeout-ms: 60000

Let each endpoint declare its own timeout. refs OAI/OpenAPI-Specification#695