Adapt masking module for LIME timeseries

[geek-yang]

Current masking module requires p_keep, which does not exist in LIME. We need to explore how LIME does masking in this case and adapt the masking module to LIME timeseries.

This issue can be addressed when issues #514 and #546 are complete.

[geek-yang]

The original LIME paper shows that the perturbation strategy for images and texts is to randomly hide one feature (e.g. one super pixel or one word bag). In my case, this could be translated to a strategy of randomly hiding one segmentation (which is adopted in the lime-for-time implementation as well, see https://github.com/emanuel-metzenthin/Lime-For-Time/blob/3af530f778ab2593246cefc1e5fdb28fa872dbdf/lime_timeseries.py#LL185C28-L185C45).

Therefore, we need to modify the masking function and make sure that in each instance one feature/segmentation will be masked.

[geek-yang]

Suggested by @cpranav93 in PR #578, we can have the "masks[0,:,:] = 1.0" as a flagged option in maskers.py rather than in lime. This can be addressed when adapting masking functions to LIME timeseries.