Question- data skewness and transformation

I have a sample size of 4290
The outcome variable (physical activity) is highly skewed and does not change with any form of transformation.
I need to model it as continuous to capture small benefits rather than categorically. Can I still do multiple linear regression since my sample size is not so small?
Or can I categorise physical activity (no, low, moderate, high) and perform ordinal logistic regression?


We have found that a semiparametric model such as the proportional odds ordinal logistic model works well for physical activity data. Among other advantages it handles lots of zeroes. Resources about this model are here.

Thanks for the very informative direction!