Keep the commit id in archive names for Windows drive paths - #554

Reviewed the branch and ran the touched tests locally — 18 passed. The diagnosis in the description is right, and stripping the drive before joining does restore the commit id. The remaining concern is that the fix is still expressed in terms of the host's path rules, so on a real Windows machine the archive entry keeps backslashes and the server would again fail to read the commit id off the first component. Inline suggestions replace the ntpath helper with one that picks a pathlib pure flavor from the running OS and emits POSIX components — that covers the separator problem, the lost UNC share, and the colon-in-a-POSIX-name false positive in one place.

Four things that have no diff line to hang off:

1. The shared OS constant the suggestions import. There are ~8 ad-hoc platform.system() / sys.platform checks scattered around the codebase today (host_info.py, ides/cursor.py, cycode_client_base.py, …). One constant in cycode/cli/consts.py, which has no imports of its own and is already imported by cyclient:

class OperatingSystem(str, Enum):
    WINDOWS = 'windows'
    MACOS = 'macos'
    LINUX = 'linux'


def _detect_operating_system() -> OperatingSystem:
    if sys.platform == 'win32':
        return OperatingSystem.WINDOWS
    if sys.platform == 'darwin':
        return OperatingSystem.MACOS
    return OperatingSystem.LINUX


OPERATING_SYSTEM = _detect_operating_system()

Migrating the existing call sites onto it is out of scope for this PR.

2. get_path_by_os in cycode/cli/apps/scan/scan_result.py:45 becomes dead weight. It flips the server's / to os.sep immediately before normalize_file_path splits the path apart again. Harmless but confusing once both sides normalize to POSIX — worth dropping the call.

3. The Windows box in the test plan is still unchecked. Given the separator point above, that is the box that decides whether the ticket is actually fixed. The suggested tests exercise the Windows rules on Linux CI by passing the flavor explicitly, which is the gap that let this reach a customer in the first place — but one manual run on Windows is still worth doing before merge.

4. Tests for path_utils now live in two places. concat_unique_id is covered in tests/test_zip_file.py while the new normalization cases sit in tests/cli/utils/test_path_utils_normalize.py. Both sets now exercise the same helper, so one home for them would be easier to keep in sync.

One style note on the suggestions themselves: the docstrings in the diff narrate the incident (used to archive the file as 'C:/repo/file', the server then reported 'C:'). That reads well today and badly in a year — the suggestions keep one line on the contract and leave the story to the commit message.