Keeping the stations in conditioned GMFs

[micheles]

Solves the case when the exposure is on top of a station, as reported in https://groups.google.com/g/openquake-users/c/pagmJS3pAyE/m/Bnn8-LLtAQAJ (reduced and converted into the test case_19). Also simplified the test scenario_risk/conditioned by using less sites and less GSIMs.

NB: since we are adding sites (the stations) the size of the correlation matrices will be different and generated GMFs different. The risk numbers are expected to change a lot in the tests with few assets and events, a little in realistic calculations. For instance for the Chi-Chi 1999 Earthquake I get

# master
| gsim                 | weight | contents      | nonstructural  | structural     | occupants | structural+nonstructural+contents |
|----------------------+--------+---------------+----------------+----------------+-----------+-----------------------------------|
| '[ChaoEtAl2020Asc]'  | 0.5000 | 3_560_892_416 | 6_912_902_656  | 5_598_780_928  | 1_586     | 16_072_576_000                    |
| '[PhungEtAl2020Asc]' | 0.5000 | 6_359_148_544 | 13_059_134_464 | 10_304_517_120 | 3_857     | 29_722_804_224                    |
| Average              | 1.0000 | 4_960_020_480 | 9_986_018_560  | 7_951_649_024  | 2_721     | 22_897_690_112                    |
# this branch
| gsim                 | weight | contents      | nonstructural  | structural     | occupants | structural+nonstructural+contents |
|----------------------+--------+---------------+----------------+----------------+-----------+-----------------------------------|
| '[ChaoEtAl2020Asc]'  | 0.5000 | 3_574_017_792 | 6_912_419_840  | 5_647_845_888  | 1_616     | 16_134_284_288                    |
| '[PhungEtAl2020Asc]' | 0.5000 | 6_297_827_328 | 12_893_349_888 | 10_241_169_408 | 3_789     | 29_432_346_624                    |
| Average              | 1.0000 | 4_935_922_560 | 9_902_884_864  | 7_944_507_648  | 2_702     | 22_783_315_456                    |