Handle excluded sessions in subject summaries

[celefthe]

Closes #27. visiomode-analysis subject now recognises sessions whose trials file carries an exclude suffix after .csv (e.g. ..._trials.csv.ignore). By default these sessions are kept in the subject summary without metrics, so they still count towards task_session. --no-ignore drops them altogether.

Changes

• The exclude suffixes are .ignore, .ignored, .exclude, .excluded and .skip, matched in any case (EXCLUDE_SUFFIXES, checked by is_excluded()). collate_sessions() now globs *trials.csv* and skips files with any other suffix (e.g. .bak).
• By default, an excluded session becomes a row with only its metadata (animal_id, session_date, protocol, environment, experiment, read from the file's first row) and every metric set to NaN. The file is never passed to session.summary(), so a broken or partial session can't stop the run. These rows still count towards session_id and task_session.
• A new excluded column in the summary CSV (right after the metadata columns) flags these rows. It is always present and is all False with --no-ignore.
• collate_sessions() and preprocess_subject() take a new ignore: bool = True parameter. ignore=False (subject --no-ignore) drops excluded sessions before ranking, and raises FileNotFoundError if no sessions are left.
• Tests cover: metadata-only rows, excluded files never being summarised, task_session counting, each suffix in mixed case, other suffixes being skipped, dropping with --no-ignore and the error when nothing is left, plus two command-line tests.
• Added a changelog entry under ## [Unreleased].