[ENH] Rethink the mask parameter in similarity search

[baraline]

Describe the issue

Currently, a boolean mask is created to indicate which part of the input time series dataset is to be ignored from the computations if the query used in the search is included in the input time series dataset.

The issue is that the computations linked to these areas are still made, they are only removed from the result array before returning it.

Suggest a potential alternative/fix

Rethink the mask parameter to be able to skip computation that won't be used instead of ignoring them later on.

Additional context

No response