Curifactory

Added
- Command flag to regenerate the report index, useful for when importing run
from another machine. (run via `experiment reports --update`.)
- Add experiment mapping step before execution - steps through experiment code
without executing stage bodies and records a list of all records and stages
they call.
- Warn if calling a stage from within another stage - this breaks experiment mapping.
- Rich library dependency, terminal logging is now fancy with colors and
progress bars!
- Logging notification if distributed run detected.
- Display control flags: `--no-color`, `--quiet`.

Changed
- Improved CLI help messages.

Added
- Paths returned from record's `get_path` and `get_dir` are now tracked and copied
into a store full run.
- `PandasCSVCacher` and `PandasJSONCacher` argument dictionaries to pass into pandas
to/read calls.
- Dirty git worktree warning in output log and indicator to output reports.

Changed
- Args hashes are now set from within the record constructor to avoid edge cases
where hashes changed and broke aggregate hashing.
- Aggregate combo hashes are set on the record directly now, done to track a
combo hash throughout a record's lifespan that started with an aggregate,
without breaking the aggregate record's potential args hash as well.

Fixed
- `PandasCSVCacher` `read_csv` not appropriately handling index column by default.

Added
- Project `curifactory` command tests.

Changed
- Project init .gitignore handling to check for/add a blank line before adding
the curfiactory section.

Fixed
- Notebook experiment folder not being created on init and from experiment.
- Notebook path config value not being used on notebook write.

Added
- Newsgroups example experiment code (see `examples/minimal/experiments/newsgroups.py`.)

Changed
- Parallel mode will automatically be set in a distributed pytorch scenario for
all processes that aren't local/node rank 0.

Fixed
- Parallel mode causing crash if global args indices are not specified.

Changed
- Minimum python version to 3.8.

Fixed
- Parallel runs with reportable caching crashing due to attempts to pickle
references containing `multiprocessing.Lock` instances.

Added
- `FileReferenceCacher` for storing lists of referenced file paths without keeping
file contents in memory.
- Automatic reportable caching. Reportables of Stages that short-circuit will now
reload and display in the report.
- Misc example project folder for experiments demonstrating various curifactory
features.

Changed
- Improve getting started documentation.
- Cacheables are now given a copy of the current record by the stages. This can
be used to access the current argset and even directly get record state within
the save/load implementation.

Fixed
- Missing files in git for example projects.