fix(windows): utf-8 subprocess decode across the desktop backend — salvage #61978 + widen to sibling sites

[teknium1]

Summary

Windows desktop backend no longer crashes with UnicodeDecodeError: 'charmap' codec can't decode byte 0x90 in subprocess._readerthread — every text-mode subprocess capture in the desktop-backend process now decodes as UTF-8 with lossy fallback.

Root cause: on Windows, subprocess with text=True and no explicit encoding= decodes child output with the ANSI code page (cp1252/GBK/…). cp1252 defines 251 of 256 bytes, so the bug stayed latent for months — until commit 84db32484f put a 🐛 emoji (UTF-8 f0 9f 90 9b) in a commit subject on main. Byte 0x90 is one of only five bytes undefined in cp1252, and the desktop update panel's git log HEAD..origin/main (_recent_upstream_commits) started pulling it through the stdlib reader thread on every Windows install behind that commit, crashing the backend (#52649, #53137, #63686, #61595, #58385).

Salvages PR #61978 by @Sahil-SS9 (server.py hardening + regression tests, cherry-picked with authorship preserved) and widens the fix to the sibling sites the crash actually came through.

Changes

• tui_gateway/server.py (salvaged from fix: harden tui_gateway subprocess reads against Windows locale UnicodeDecodeError (#53137) #61978): encoding="utf-8", errors="replace" on the slash-worker Popen, pdftoppm, cli.exec, quick-command exec, and shell.exec — plus regression tests pinning the kwargs
• hermes_cli/web_server.py: git-log update panel (the site that fired), memory-provider setup runner, WhatsApp bridge npm install, docker probe
• hermes_cli/banner.py: all 5 git sites (startup update check runs in this process)
• tui_gateway/host_supervisor.py: build-SHA probe, ps probe, compute-host Popen drain threads
• tui_gateway/compute_host.py: build-SHA probe, ps rss probe

errors="replace" is the load-bearing half: even if a child ever emits genuinely non-UTF-8 bytes, decoding degrades to U+FFFD instead of raising in a daemon thread.

Validation

Check	Result
Legacy path repro: cp1252 decode of the real trigger line	raises 'charmap' ... byte 0x90 as reported
E2E: real git repo with 🐛 commit through _recent_upstream_commits + banner._git_stdout	decodes correctly, kwargs verified
Invalid byte 0x90 through the new path	degrades to U+FFFD, no exception
tests/tui_gateway/ + banner tests	457 passed, 0 failed

Closes #61978 (salvaged with credit). Fixes #52649, #53137 (and duplicates #63686, #61595, #58385).

Infographic

[github-actions]

૮ >ﻌ< ა ci review

_{ran on a618210}

all good!