Masking the Gaps: An Imputation-Free Approach to Time Series Modeling with Missing Data