Pycmtensor

Feat

- **functions.py**: added `gelu` function

Fix

- **basic.py**: fix bug in `null_log_likelihood_fn` for givens
- **dataset.py**: fixed slice call when no `batch_size` argument is used
- **dataset.py**: fixed the `split` method to accept `None` as argument

Refactor

- **functions.py**: refactored method to covert list of utilities to tensor variables
- **dataset.py**: accept mixed `str` and `TensorVariable` types in `_make_tensor`
- **basic.py**: moved `build_gh_fn()` to `BaseModel`

Fix

- **pyproject.toml**: fix incorrect bump message

Feat

- **run**: added `is_training` flag

Fix

- added `null_log_likelihood_fn` to basic.py
- **results**: add argument for max cutoff in `show_training_plot`
- **results.py**: fixed legend placement on training plot
- **run.py**: add option for acceptance_method argument

Refactor

- minor refactorization

Feat

- **layers.py**: added new layer type `LayerNorm`
- **expression.py**: added new parameter type Gamma
- **results.py**: add argument offset to show_training_plot
- **run**: moved train and compute to pycmtensor.run for cleaniness

Fix

- **TasteNet.py**: changed instance check type from .DenseLayer to .Layer
- **run.py**: add limit for verbosity

Fix

- **dataset**: added split via n-count and added option for pre-shuffle
- **dataset**: updated scale_variable method to handle both single variables and lists of variables

Fix

- **workflow**: remove poetry from tests
- add missing poetry dependencies