TMGimporter: skip real-DB import tests on Windows

[eduralph]

Root cause

The import tests in TMGimporter/tests/test_libtmg.py drive a real in-memory
Gramps database (make_database + db.load) through the shared _make_db()
helper. On Windows the only Gramps available to CI comes from conda-forge, which
has no 6.1 yet and where Gramps 6.1 does not build (its Windows build targets
MSYS2 UCRT64, not conda); run against that mismatched Gramps the real-DB import
path hangs. The Windows CI lane that surfaces this is the one introduced by
#820.

Fix

Raise unittest.SkipTest from _make_db() when sys.platform == "win32". Every
real-DB test obtains its database through this single helper (all 43 db
creations route through it), so the 13 DB-backed classes skip in one place. The
pure-function tests (code stripping, date conversion, name parsing) do not call
_make_db() and continue to run on Windows.

Verified against

• TMGimporter/tests/test_libtmg.py — _make_db() is the sole make_database
  call site; confirmed every DB-backed class reaches the database only through
  it, so one guard covers them all.
• Linux: the full suite runs unchanged — 157 tests pass.
• Windows path (with sys.platform forced to win32): 95 tests skip, 62 run
  and pass, and the module is not all-skipped (the pure-function classes keep
  running), so a skip-detecting runner does not mistake it for an empty module.

Test

This change is itself in the test suite. Pre-change, the Windows lane added by
#820 hangs in import_* against the mismatched Gramps; post-change it skips the
real-DB tests cleanly. Validated on that lane (gramps61): the module reports
157 tests, 95 skipped, 0 broke in ~5 min (previously an indefinite hang), with
the Linux run unaffected. The skip self-heals once Gramps 6.1 reaches conda-forge
and the Windows lane runs the matching Gramps.

[eduralph]

I will have to see how to address this issue once the CI process has landed.

[GaryGriffin]

Is there a MacOS lane for CI?

[eduralph]

@GaryGriffin - I looked into that and didn't bother so far because we can't build a ui test and there's a heavy overlap with linux on the unit test level. I can add if you like - gramps does have one for mac specfic tests

[jralls]

I think I'd need to build an installation tarball of the dependencies. It would take forever to build everything from scratch and I don't know if Homebrew has everything we need.