M6anet

- Add --pretrained model flag for users to specify m6Anet models to be run
- Add m6Anet model trained on RNA004 kit
- Modify README to include installation and quickstart intructions

- Relax package requirement for m6anet
- Modify documentation to install using pip from source
- Add pytest-dependency custom mark for dataprep test

Github action on on publishing v-2.0.0 was run wrongly, resulting in users not being able to install the package from PyPI. This minor release fixes the issue

- Single entry point for all m6Anet functionalities such as m6anet dataprep, m6anet inference, and m6anet train
- Faster inference procedure by first saving individual read probability on data.indiv_proba.csv before sampling the required number of reads and averaging the calculated site probabilities
- Dataprep option to round data.json output to 3 decimal places
- Provides m6Anet model trained on Arabidopsis VIRc dataset from `https://elifesciences.org/articles/78808` and the corresponding normalization factors

- Fixed torch version to prevent failed installation when latest version of pytorch is installed in a machine without GPU
- Fixed wrong naming convention with training demo data data.readcount.labelled
- Filter warning messages from pandas for less cluttering
- Fix typo m6anet-run_inference --infer_mod_rate flag (it is now --infer_mod_rate instead of --infer_mod-rate)

- Output kmer motif on the result file from m6anet-run_inference
- Add an option to automatically filter segments with low number of reads during m6anet-dataprep to reduce output size
- Add new functionality for pooling of reads from multiple replicates during training and inference
- Add new functionality for single molecule stoichiometry prediction
- Update to documentations