fix(configure-resources): handle Windows (Git Bash) host detection

[mpasternak]

Problem

The CI job Testy Makefile (Windows) has failed on every push to main since the resource-limits redesign (#5), which added a test that runs scripts/configure-resources.sh on windows-latest. main has been red since 2026-06-01 (visible on the #5 and #7 merge runs alike — 65 passed, 1 failed).

Root cause: the script's host detection only matched Linux / Darwin from uname -s. On Git Bash uname -s is MINGW64_NT-..., which falls through to the *) branch → exit 1:

TEST: configure-resources — staly cap + uslugi zmienne
  FAIL: configure-resources zwrocil blad
  RESULTS: 65 passed, 1 failed, 2 skipped

Unrelated to the media-root change in #7 — that PR just inherited the pre-existing red.

Fix

Add a MINGW*|MSYS*|CYGWIN* branch to the uname -s case:

• RAM from /proc/meminfo (emulated by the MSYS runtime on Git Bash), CPU from nproc.
• Graceful fallbacks so it never exits non-zero: CPU → NUMBER_OF_PROCESSORS → 1; RAM unreadable → TOTAL_RAM_GB=0, and the existing 4 GB budget floor takes over.
• Fixed caps (redis, netdata, …) are host-independent, so .env still gets correct values even in the worst case.

BPP deploys on Linux; Windows is a dev-only environment, so finishing and writing .env is all that's required there.

Verification

• Worst case simulated locally (uname stubbed to MINGW64, no /proc/meminfo, NUMBER_OF_PROCESSORS=4): script exits 0 and produces every value the CI test asserts — REDIS_MEM_LIMIT=1024m, …=192m/128m/320m/64m, REDIS_MAXMEMORY=819mb, presence of APPSERVER/DBSERVER/WORKER_GENERAL_MEM_LIMIT + DBSERVER_CPU_LIMIT, and absence of FLOWER_CPU_LIMIT.
• macOS native path unchanged (64 GB / 16 CPU detected, fixed caps correct).
• bash -n clean; pre-commit incl. shellcheck green.

🤖 Generated with Claude Code