Less broadcast for numpy exercise

Currently the indexing in the numpy example and exercise revolve a lot around broadcasting.

This is not necessary very useful. Other kind of indexing, like selection with boolean mask or range selection may be more interesting.

Locate the numpy example and exercise referenced in the issue and compare how much they rely on broadcasting. Revise the material to emphasize boolean-mask selection or range selection instead, then check that the example and exercise remain coherent for the workshop.

Written by the indexing model from the issue text.