fix(windows): suppress git-probe console flash + cp950 crash in dashboard

[kenchan-pixel]

What does this PR do?

On Windows, tui_gateway/git_probe.py::run_git spawned git with no creationflags and no explicit encoding. Two consequences:

1. Console-window flash per repo. The gateway/dashboard runs windowless (pythonw.exe), so each bare git.exe allocates a fresh console (conhost.exe, and with Windows Terminal as default terminal an OpenConsole.exe window). The Projects-tree scanner calls run_git several times per project directory, so launching the desktop app flashes dozens of console/terminal windows for users with many repos. (Measured on one affected machine: ~118 process spawns in 60s at startup, almost all git -C <dir> rev-parse ..., each spawning a conhost.)
2. UnicodeDecodeError on non-ASCII repo paths. With text=True and no codec, the reader thread decodes git's UTF-8 output with the locale ANSI codepage (cp950 on Traditional-Chinese Windows) and crashes on any repo whose path/branch contains non-ASCII (e.g. CJK) characters, spamming gateway-crash / _readerthread logs.

Fix at the single run_git choke point: add encoding="utf-8" / errors="replace" and creationflags=windows_hide_flags() (the existing hermes_cli/_subprocess_compat.py helper — CREATE_NO_WINDOW on Windows, 0 elsewhere, so it's a no-op on macOS/Linux).

Related Issue

Fixes #53178

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• tui_gateway/git_probe.py: run_git now passes encoding="utf-8", errors="replace" and creationflags=windows_hide_flags(); import the helper from hermes_cli._subprocess_compat.
• tests/tui_gateway/test_git_probe.py (new): asserts the call wiring (no-window flag + utf-8/replace), empty result on non-zero exit, and no spawn for empty cwd. Cross-platform (mocks subprocess.run).

How to Test

1. Repro (before): On Windows (default terminal = Windows Terminal) with ~15–20 git project dirs (a couple with CJK paths), launch the desktop app and watch git.exe → conhost.exe → OpenConsole.exe windows flash; logs show _readerthread UnicodeDecodeError: 'cp950'... for the non-ASCII repos.
2. After: 8 run_git/branch probes (incl. a CJK-path repo) from a pythonw parent → 0 new console/OpenConsole windows and no decode crash (returned correct UTF-8 toplevel + branch).
3. tests/tui_gateway/test_git_probe.py → 3 passed. scripts/check-windows-footguns.py → clean.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(windows): ...)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix (no unrelated commits)
• I've added tests for my changes
• I've tested on my platform: Windows 11 (cp950 locale). The change is a no-op on macOS/Linux (windows_hide_flags() returns 0; explicit utf-8 is safe everywhere).

Note: I ran the new test file (pytest tests/tui_gateway/test_git_probe.py → 3 passed) and scripts/check-windows-footguns.py (clean), not the entire suite locally — CI will run the full suite.

Documentation & Housekeeping

• Documentation — N/A
• cli-config.yaml.example — N/A
• CONTRIBUTING.md/AGENTS.md — N/A
• Cross-platform impact considered (no-op on non-Windows)

[teknium1]

Closing as superseded by the consolidated Windows console-flash work tracked in #54220.

The relevant pieces from this PR/cluster have now landed through the targeted follow-up PRs #54236, #53892, and #54417, or are recorded in the umbrella tracker for any remaining native-Windows verification. Keeping this separate PR open would duplicate the tracker and the merged follow-up work.

Thanks for digging into this — the reports and PRs in this cluster helped identify the remaining spawn legs.