`InvertedIndexVectorized._set_logger` shares a process-wide logger across all dataset instances

[heathersherry]

Problem

InvertedIndexVectorized._set_logger (in muller/core/query/inverted_index_vectorized.py) does:

logger = logging.getLogger('my_logger')
...
if not logger.handlers:
    file_handler = logging.FileHandler(path, mode='a')
    ...

Because Python's logging.getLogger returns a process-wide singleton keyed by name, and the if not logger.handlers guard short-circuits on subsequent calls, the FileHandler bound during the FIRST InvertedIndexVectorized(...) construction is reused for the entire process. All later instances—even those for a different dataset path—log to that first file.

Why it matters

• Cross-dataset log mixing in any workflow that opens more than one dataset (e.g. tests, batch jobs, notebooks that load several datasets).
• The "remote path → console-only" branch added in Phase 1 only takes effect if the remote dataset is the FIRST one constructed in the process; otherwise it silently inherits a stale local FileHandler.
• The logger name 'my_logger' is generic enough to collide with any other library that uses the same name.

Suggested fix

Use a per-instance child logger keyed by dataset path / column, e.g.:

name = f"muller.inverted_index.{_safe(self.dataset.path)}.{self.column_name}"
logger = logging.getLogger(name)

Or accept that this is best-effort logging and just use the standard module-level logger (logging.getLogger(__name__)) with NullHandler, letting application code configure handlers. The FileHandler at a dataset-derived path is the surprising part; consider dropping it entirely if the same information is available via the module logger.