fix(stack): pool gateway writes on keep-alive connections (CLI-2623)

[7ttp]

TL;DR

stops the local stack gateway from opening a new upstream connection for every write.

whats broken?

every .rpc(), auth.admin.createUser, delete or upload went upstream on its own connection and left it in TIME_WAIT, so a busy test run used up the macOS port range.

now fixed by:

writes reuse the gateway's keep alive pool like reads do. a write is still never resent.
Studio and its /mcp route keep a fresh connection per write, as before

chunked GET and DELETE bodies are forwarded as chunked, so they reach the upstream and stay off the next request
PS: Auth's own Postgres churn is tracked in #6981

ref:

• closes: stack: gateway still opens a new upstream connection per request on 2.119.0 (exhausts ephemeral ports on macOS) #6987
• extends: fix(stack): reuse keep-alive upstream connections in the HTTP gateway #6925

[github-actions]

🤖 AI Review

Both independent reviews were available. Claude's finding is confirmed as a minor reliability trade-off: pooled writes can fail on a stale connection without retry. Studio's fresh-connection exception is correctly propagated. No additional findings were identified. Tests were inspected but not run.

Findings

Severity	Location	Category	Sources	Claim
🟡 MINOR	packages/stack/src/HttpProxy.ts:433	reliability	claude	Writes on routes without freshWrites now reuse pooled upstream connections. If an upstream closes a connection as it is reused, the write fails with 502 without retry. Previously, writes used fresh connections and avoided this particular failure mode. The four-second agent timeout mitigates idle-timeout races but cannot eliminate connection-reset races.

Stats

Claude findings: 1 · Codex findings: 0 · Confirmed: 1 · Refuted: 0 · Uncertain: 0

---

Models: claude-opus-5-5 + gpt-6.1-sol · Trigger: auto · Workflow run

This review runs once per PR. A maintainer can request another with a /ai-review comment.