fix(pool): correct turn management in putIdleConn to prevent connection leaks

[cyningsun]

Bug Description

A critical defect was identified in the connection pool's turn management mechanism within the putIdleConn method, breaking the fundamental 1:1 correspondence between turns and connections. ( Introduced by #3518)

Root Cause Analysis

Turn Release Scenarios (Expected Behavior):

- new connection failed
  → freeTurn: allow other goroutines to create connections
- pool.Put 
  → freeTurn: allow acquiring existing idle connections  
- pool.putIdleConn
  ├── deliver connection to waiting goroutine
  │   → ❌ DEFECT: turn should be released by receiving goroutine via Put
  └── place connection in idle pool
      → freeTurn: allow acquiring existing idle connections

The Defect: In putIdleConn, when delivering connections to waiting goroutines, the turn was incorrectly released immediately instead of being transferred to the receiving goroutine for proper release via Put.

Test Gap: The test case "should not leak turn when delivering connection via putIdleConn" contained flawed assertions due to misunderstanding of turn ownership mechanism in the context of connection delivery, failing to detect the leakage.

Relationship to #3625:
This PR addresses the same turn management issue identified in #3625. I want to acknowledge and thank @ndyakov for:

• The comprehensive test cases that clearly reproduce the issue
• The prompt response in providing a solution
• The valuable implementation in fix(pool): double freeturn bug #3625

This PR offers an additional perspective on the solution, focusing specifically on turn ownership semantics. It's intended to provide more options for review and discussion as we work towards the best resolution for this issue.

Note to Reviewers:
This PR is optional and can be closed if #3625 covers all needs. It's just here to offer another perspective.

[jit-ci]

Hi, I’m Jit, a friendly security platform designed to help developers build secure applications from day zero with an MVS (Minimal viable security) mindset.

In case there are security findings, they will be communicated to you as a comment inside the PR.

Hope you’ll enjoy using Jit.

Questions? Comments? Want to learn more? Get in touch with us.

[Copilot]

Pull request overview

This PR fixes a critical bug in the connection pool's turn management system that could cause turn leaks and incorrect semaphore accounting. The fix ensures proper transfer of turn ownership when connections are delivered to waiting goroutines.

Key changes:

• Modified putIdleConn to return a boolean indicating whether the caller should free the turn
• Updated queuedNewConn logic to conditionally free turns based on putIdleConn's return value
• Corrected test assertions to reflect the proper turn ownership semantics

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
internal/pool/pool.go	Modified putIdleConn to return boolean for turn management, refactored queuedNewConn error handling to properly free turns only when connections are added to idle pool rather than delivered to waiting goroutines
internal/pool/pool_test.go	Updated test expectations to correctly validate that turns are held during connection delivery and only freed after the receiving goroutine completes its own dial or returns the connection via Put()

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[ndyakov]

I do personally like this approach better, it is easier to follow and read. Thank you @cyningsun. Will let @ofekshenawa decide which one to include. Could you please sync the rest of the changes (the context calculation and the tests).

[cyningsun]

@ndyakov, thank you! I'm glad you like the approach. I've gone ahead and synced the context calculation and tests from #3625. All tests are passing.