fix: Windows breakage after #16 — sys.path shadow and Microsoft OpenSSH multiplex

[battlesnake]

Two issues that hit Windows users after #16:

1. mama_ssh.py shadows stdlib types

When git invokes the wrapper as python mama_ssh.py …, __package__ is empty and the script took a fallback path:

sys.path.insert(0, os.path.dirname(_THIS_DIR))   # = .../mama
from utils import ssh_multiplex

That puts <...>/mama/ at the front of sys.path. Inside mama/ lives a types/ subpackage (mama.types — git/asset/dep_source helpers), which then shadows Python's stdlib types for any subsequent first-time import. The next from types import MethodType, GenericAlias (e.g. inside contextlib, transitively imported by ssh_multiplex.py) explodes:

ImportError: cannot import name 'MethodType' from 'types'
(...site-packages\mama\types\__init__.py)

It mostly bit uv-tool-installed Pythons, where contextlib hadn't been pre-cached before our path manipulation ran. Fix: try the qualified from mama.utils import ssh_multiplex first (works for any pip-installed mama with no path manipulation needed); fall back to inserting the package's parent — never mama/ itself.

2. Microsoft OpenSSH for Windows ControlMaster is flaky

Masters drop with "Connection reset by peer" mid-session and the stale socket then blocks reattach:

mux_client_request_session: read from master failed: Connection reset by peer
ControlSocket C:\Users\jorma/.ssh/cm\<hash> already exists, disabling multiplexing

Detect the buggy client narrowly by ssh -V banner — only the Microsoft port prints OpenSSH_for_Windows_<ver>. When that's the active ssh, skip multiplex; keepalives, parallel fetches, and the concurrency cap stay enabled. Cygwin, Git-Bash/MSYS2, and WSL ssh on Windows report the standard OpenSSH_<ver>p1 banner and keep full multiplex. Result is cached per process.

User config (~/.ssh/config) is never overridden in either case.

[battlesnake]

Pushed two follow-ups:

• f098b2d — switched to System.windows per your inline comment.
• 4bb6c77 — refined further: instead of disabling multiplex on all of Windows, probe ssh -V at startup and only skip when the banner contains for_Windows (Microsoft's port). Cygwin, Git-Bash/MSYS2, WSL ssh on Windows all report the standard banner and have working ControlMaster, so they get full multiplex. Result is cached.

Also tightened the inline comment, fixed an inaccurate docstring ("unsupported" → "known-broken"), and added a test for the "Windows + user has multiplex configured in ssh config" path. 41/41 tests pass.

PR description updated to match.

[battlesnake]

Pushed 18dabd0 — fixes a second Windows-only bug from #16 that surfaced in Slack today (Mykhailo):

ImportError: cannot import name 'MethodType' from 'types'
(...site-packages\mama\types\__init__.py)

Reproduced and verified locally. The wrapper's fallback path inserted <...>/mama onto sys.path, so mama/types/ shadowed stdlib types the moment something (here: contextlib, transitively via ssh_multiplex.py) did from types import MethodType. uv-tool-installed Pythons don't pre-cache types, so the shadow triggered.

Fixed by importing as the qualified from mama.utils import ssh_multiplex; only fall back to sys.path manipulation when even that fails, and add the grandparent of mama_ssh.py (so mama/ is reachable as a package without leaking mama/types/ onto the path).

Added a regression test that runs the wrapper in a fresh subprocess and asserts the mama/ dir does not appear on sys.path. 42/42 tests pass. PR title + description updated to cover both fixes.

Note: mama.types/ shadowing stdlib types is a pre-existing footgun in mama itself — long-term it might be worth renaming to mama.dep_types or similar, but that's out of scope here.

[battlesnake]

Continued in #19 (auto-closed here when the source fork was deleted; same 5 commits at 18dabd0 now sourced directly from RedFox20:feature/supermama).