fix(task): prevent hang when tasks with multiple dependencies fail

[stempler]

Summary

Fixes task execution hanging when a task depends on multiple tasks that share a common failing dependency. Extends the fix from #6430 to handle the main scheduler loop.

Problem

When multiple tasks depend on a common dependency that fails (e.g., lint, test, and build:docker all depend on install, which fails), and another task depends on all of them (e.g., check depends on all three), the scheduler would hang indefinitely instead of properly exiting with an error.

This scenario wasn't covered by the previous fix in #6430, which only handled nested task sequences within inject_and_wait.

Root Cause

In the main scheduler loop:

1. When a dependency fails, is_stopping() becomes true
2. The scheduler stops accepting new tasks from the queue
3. However, remaining tasks stay in the main_deps dependency graph
4. The spawned async task waits for main_deps to become empty before sending the main_done signal
5. This creates a deadlock - the scheduler has stopped but the deps graph is never cleared

Solution

Added cleanup logic similar to inject_and_wait: when is_stopping() is detected in the main scheduler loop, explicitly remove all remaining tasks from main_deps. This triggers the completion signal and allows proper shutdown.

Test plan

• Added e2e test test_task_failure_hang_depends that reproduces the complex dependency scenario
• Test verifies that mise exits with error code 1 instead of hanging when dependencies fail
• Existing test test_task_failure_hang continues to pass
• All task-related e2e tests pass

Fixes: #6391

🤖 Generated with Claude Code

Co-authored-by: Claude noreply@anthropic.com

[Copilot]

Pull Request Overview

This PR adds a reproduction test case for a reported issue where mise hangs when tasks fail with dependencies. The test validates that the task runner properly exits with failure instead of hanging indefinitely when dependent tasks fail.

• Adds an end-to-end test script to reproduce the hanging behavior described in the linked GitHub discussion
• Creates a test scenario with multiple interdependent tasks where failures should propagate without causing hangs
• Uses timeout mechanism to detect and fail the test if hanging occurs

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[jdx]

thanks for the help @stempler—hopefully we've got these issues sorted now