fix: support uv venv backend under MSYS2/MinGW (#1088)

[henryiii]

Okay, I think this is a valid fix for #1088. @Tiphereth-A, would be good if you could take a look to be sure. :)

🤖 AI text below 🤖

Fixes #1088.

Problem

Under MSYS2/MinGW, uv cannot inspect the native interpreter — it reports an unknown operating system (e.g. mingw_x86_64_ucrt_gnu). Because script mode runs uv venv -p <sys.executable> ..., venv creation failed even though a bare uv venv works fine.

Fix

• In VirtualEnv.create, skip -p for the uv backend on MinGW when no interpreter is explicitly requested, letting uv pick the default (which works). An explicit interpreter is still passed through, and non-MinGW behavior is unchanged.
• uv builds a Windows-style venv (Scripts/python.exe) even under MinGW, whereas the native interpreter and virtualenv/venv use a POSIX layout (bin/python). A new _windows_layout property centralizes this so bin_paths and the environment-reuse check select the right directory per backend.

Based on the patch from @Tiphereth-A in msys2/MINGW-packages#28759.

Tests

Added tests for the MinGW uv layout, the MinGW no--p command, and the explicit-interpreter uv path. nox/virtualenv.py and tests/test_virtualenv.py pass lint and the full test_virtualenv.py suite.

[henryiii]

🤖 AI text below 🤖

Pushed one commit adding MinGW-host test corrections (test_bin_paths, test_bin_windows, test_create): on a real MSYS2/MinGW host sys.platform == "win32" but the native venv uses a POSIX bin/ layout with .exe executables, so the Windows-layout assertions must exclude MinGW. These are no-ops off MinGW (the existing matrix stays green) and only matter under the experimental MSYS2 job in #1122.

For context: #1122 stands up that MSYS2 job and acts as a real-host demonstration of this PR. Without this PR, #1122 shows the #1088 symptom directly — uv venv -p .../bin/python.EXE → Failed to inspect Python interpreter — across test_uv_creation, test_create_reuse_uv_environment, test_stale_environment[uv-virtualenv-False], and the test_noxfile_script_mode* trio. With this PR merged, that job goes fully green (726 passed). The two PRs are independent diffs and can land in either order.

[Tiphereth-A]

LGTM

[henryiii]

🤖 AI text below 🤖

Correction to my earlier comment: the MinGW-host test corrections (test_bin_paths, test_bin_windows, test_create) have been moved to #1122 instead — they cover the native bin/ layout that predates this PR (#901), so they belong with the CI job that exercises them. This PR now contains only the uv-backend fix and its own tests.

The relationship is unchanged: #1122 stands up the MSYS2 job and demonstrates this PR on a real host. With its own corrections in place, #1122's only remaining red is the uv path — uv venv -p .../bin/python.EXE → Failed to inspect Python interpreter (the #1088 symptom) — which this PR fixes. The two PRs share no commits and can land in either order.