Refactored conditioned GMFs a bit more

[micheles]

Also improved the monitoring to measure the slow spots. Part of #9003. Here is @CatalinaYepes calculation on oq1:

| calc_54926, maxmem=0.4 GB | time_sec | memory_mb | counts |
|---------------------------+----------+-----------+--------|
| total gen_event_based     | 46_522   | 34.5      | 1000   |
| reading mea,tau,phi       | 17_410   | 0.0       | 1000   |
| computing gmfs            | 4_345    | 0.0       | 1000   |
| EventBasedCalculator.run  | 210.9    | 76.8      | 1      |
| filtering ruptures        | 49.0     | 0.0       | 1000   |
| saving gmfs               | 7.36327  | 0.25781   | 1000   |

As I suspected, it can be faster to recompute mea,tau,phi on every task rather than reading such arrays.

[raoanirudh]

qa_tests_data/event_based_risk/reinsurance_1/gmfs.hdf5: stray file being committed?