fix(gateway): suppress black console window flash on Windows

[seamusmore]

What does this PR do?

Fixes the black console window flash that appears when Hermes gateway (running as on Windows) spawns child processes like bash, git, cmd, or powershell.

Root Cause

When the gateway is launched via (GUI subsystem), it has no console. Every console-subsystem child it spawns then allocates its own visible console → black window flashes.

Fix

Allocate a hidden console in so child processes inherit it instead of creating their own visible windows.

• check prevents stealing console from interactive CLI
• ensures gateway still starts if fails
• POSIX path is a no-op ()

Type: 🐛 Bug fix

Changes

• ****: +19 lines in before the duplicate-instance guard

Platform

Windows

[teknium1]

Thanks for investigating the Windows console flashes. The desktop Git Bash resolver extraction is focused and follows the existing dependency-free Vitest helper pattern.

Problems

• The stated bash path is already covered on current main: tools/environments/local.py:1211 passes windows_hide_flags() to subprocess.Popen. The detached gateway is deliberately launched as console-less pythonw.exe in hermes_cli/gateway_windows.py:883-915. Before attaching a process-wide console in gateway/run.py, please identify and reproduce a remaining unprotected child-spawn path.
• The new AllocConsole() path has no regression coverage. The added tests exercise only findGitBash.

Suggested changes

• Add a focused, injectable helper test covering existing-console, no-console, and AllocConsole-failure behavior, and validate the affected Windows child path.
• Keep the console change scoped to the demonstrated missing spawn path if per-child windows_hide_flags() is sufficient.

This is an automated hermes-sweeper review.

[teknium1]

Please identify the current child-spawn path that still flashes before attaching a console to the entire detached gateway. The normal terminal bash path already passes windows_hide_flags() on Windows (tools/environments/local.py:1211); add a focused regression test/repro for the remaining path and preserve the detached pythonw.exe lifecycle contract.

[seamusmore]

After investigation per your review suggestion, here is what I found.

Per-child approach investigation

Following your suggestion to identify "a remaining unprotected child-spawn path" instead of the process-wide AllocConsole(), I traced every subprocess spawn path in the gateway:

✅ Paths already covered (have creationflags)

File	Call	Status
tools/environments/local.py:1211	subprocess.Popen (terminal tool)	✅ windows_hide_flags()
gateway/run.py:6319	subprocess.Popen (watcher)	✅ windows_detach_flags_without_breakaway()
gateway/run.py:6354	subprocess.Popen (watcher)	✅ **windows_detach_popen_kwargs()
gateway/slash_commands.py:4612	subprocess.Popen	✅ **windows_detach_popen_kwargs()

❌ Paths WITHOUT creationflags (Windows only)

File	Call	Child
gateway/run.py:2156	asyncio.create_subprocess_exec	ffprobe
gateway/run.py:10106	asyncio.create_subprocess_shell	quick commands
gateway/platforms/qqbot/adapter.py:2115	asyncio.create_subprocess_exec	ffmpeg
gateway/platforms/whatsapp_cloud.py:1249	asyncio.create_subprocess_exec	ffmpeg
gateway/platforms/signal.py:169	subprocess.run	ffmpeg
gateway/platforms/webhook.py:1186	subprocess.run	gh pr comment

Why per-child doesn't scale

I implemented the targeted fix (adding creationflags=windows_hide_flags() to all 6 unprotected paths above + regression tests in test_windows_subprocess_no_window_flags.py), pushed the update, and tested. The console flash persisted.

The fundamental problem is that the codebase has dozens of spawn points across gateway/, tools/, and platforms/, plus third-party libraries that spawn their own children. Any path I missed — or any new one added in a future commit — will flash. The process-wide AllocConsole() is the only approach that guarantees coverage:

1. It gives the pythonw.exe gateway a hidden console that all descendants inherit, regardless of which module spawned them
2. It covers third-party code (ffmpeg, ffprobe, node, etc.)
3. It's future-proof against new spawn paths being added
4. It doesn't change the detached lifecycle — the console is hidden via ShowWindow(hwnd, 0), and the process remains a valid detached gateway

Test coverage

Added 5 regression tests in tests/test_windows_subprocess_no_window_flags.py:

1. test_async_create_subprocess_hides_window — verifies asyncio.create_subprocess_exec passes CREATE_NO_WINDOW
2. test_alloc_console_skips_when_console_exists — interactive CLI path (no-op)
3. test_alloc_console_hides_when_no_console — pythonw.exe path (alloc + hide)
4. test_alloc_console_survives_failure — AllocConsole failure → caught, gateway continues
5. test_alloc_console_noop_on_posix — POSIX is unaffected

All 20 tests pass (15 pre-existing + 5 new).

[teknium1]

The desktop-parity half of this PR is merged via PR #70265 — your find-git-bash.ts extraction, the HERMES_GIT_BASH_PATH override, and your tests landed with your commit authorship preserved (rebase-merge). The console/AllocConsole half was superseded by the class-level fix in PR #70205 (commit 0dbf639), which eliminated the console-less parents that caused the flashes at the launcher layer instead of per-site.

Thanks for both halves — the parity gap was a real find.