Being transactional is orthogonal to being immutable. For example, if you want to ensure that a given set of writes is applied atomically, you would issue them in a transaction.
This is especially important if you are also dealing with metadata that you don't want to always denormalize into your time-series tables. If you don't insert it transactionally, you can also lose referential integrity between your time-series and relational (metadata) tables.
That's not always true. There are tons of cases where time-series data gets corrected later. It is true that time-series is INSERT-mostly but UPDATES do happen.