You are absolutely right. There could be a significant interaction (moderator) effect despite the fact that the main effect is zero. For example, a drug might have a positive (beneficial) effect for older people but a negative (detrimental) effect in younger people. Then, positive and negative effects may cancel out so that the overall (main) effect of the factor "drug" is zero. Yet the moderation (interaction) effect drug*age would be non-zero and important to know about. In the literature, this is sometimes referred to as a "disordinal" interaction.

Please refer to Baron and Kenny’s approach (1986) which represented in four steps to establish a significant mediation effect as below;

First, there must be a significant relationship between the predictor and the outcome. Second, the predictor must be significantly related to the mediator. Third, the mediator should be significantly related to the outcome variable. Finally, there is a significant mediation effect when the relationship between the predictor and the outcome becomes significantly weaker (partial mediation) or non-significant (full mediation) after the inclusion of the mediator.

If you found the effect of variable X on Y is nonsignificant, you can still examine the moderation indirect effect of M on the pathways. There is sig. indirect even one (or both) pathways that define indirect effects are found insignificant. This indicates that M variable fully mediates the relationships between variable X and Y. According to Hayes (2017), the insignificant indirect effects exist as path a is multiplied to path b, regardless of whether one of the paths is insignificant to indicate M variable has sig. indirect effects.