Investigate and fix the memory problem in the EventBased risk calculator

[Michele Simionato]

We discovered in the context of a computation by Miriam (see attached files) that Python is allocating a lot of memory which is never released. The memory occupation is much bigger than the expected size of the gmvs_ruptures array. Sounds like a memory leak.

[Michele Simionato (michele-simionato)]

Here are the pull requests: https://github.com/gem/oq-engine/pull/1167 (improved the memory profiling) and https://github.com/gem/oq-engine/pull/1166 (cleaned up the hazard getters to save memory).